Journal of Public Economics - Mazhar Waseemmazharwaseem.com/static/uploads/Waseem2018_JPubE... · (IGC), Pakistan Program is gratefully acknowledged. E-mail address: [email protected]

Contents lists available at ScienceDirect

Journal of Public Economics

journal homepage: www.elsevier.com/locate/jpube

Taxes, informality and income shifting: Evidence from a recent Pakistani taxreform☆

Mazhar WaseemDepartment of Economics, University of Manchester, Arthur Lewis Building, Oxford Road, Manchester, United Kingdom

A R T I C L E I N F O

Keywords:EfficiencyIncome TaxInformalityTax evasion

JEL classification:H21H240O17H26

A B S T R A C T

This paper examines firm behavior to taxation in a low enforcement and large informality setting. Using quasi-experimental variation created by a tax reform, which increased taxation of partnerships substantially relative tofirms of other legal form, and the population of income tax returns filed in Pakistan in 2006–11, I document thattreated firms report significantly lower earnings, migrate into informality, and switch business form in responseto the increase in tax rate. The revenue loss caused by these behavioral responses is so large that by the third yearafter the reform the government was collecting less revenue than it would have without the tax increase. Thisimplies that the new tax rate was on the wrong side of the Laffer curve and would not have been optimal underany social preferences. The richness of the data and tax variation permits me to decompose the observed re-sponses into real and evasion margins and to identify fiscal externalities created by the reform on other tax bases.The welfare cost of the reform increases by around 40% once these externalities are taken into account.

1. Introduction

The presence of large informal sector constrains taxation capacity ofdeveloping countries in two important ways.1 First, there is a directeffect as taxation base is limited to a narrow set of formal taxpayers.Second and more subtle is the indirect effect: governments in thesecountries tend to keep tax rates low fearing that increased taxationmight unravel the already thin formal sector.2 Whether such fears arejustified depends on the elasticity of the tax base, in particular on howlikely the taxpayers are to exit into informality in response to a taxincrease. There is quite a large literature that estimates the sensitivity ofthe tax base to the marginal tax rate using administrative tax returndata (Saez et al., 2012), but unfortunately most of this literature is set inrich countries and the corresponding evidence for developing countriesis limited. In fact, to my knowledge there is no micro-based study thattakes into account the movements into and out of informality, whicharguably is a more important margin of response to taxation in a de-veloping country setting. This paper fills the gap by presenting evidenceon the responsiveness of earnings, formality and business organizationchoices of agents to personal income taxation in Pakistan.

For this purpose, I exploit a natural policy experiment created by an

income tax reform introduced in the country in 2009. Before the reformearnings of noncorporate firms – sole proprietorships and partnerships –were taxed lightly relative to earnings of corporations, and it was feltthat the distortion was preventing the incorporation of new firms. Thereform raised the income tax rate on partnership earnings to a flat 25%,thus neutralizing largely a partnership’s incentive to stay unin-corporated. As an unintended consequence, however, it created a largetax rate variation within noncorporate firms: partnerships experiencedon average a greater than five-fold increase in tax rates from 2009,while rates applicable to sole proprietorships remained unchanged in2009 but reduced slightly from 2010 when their tax schedule was re-vised.3 These differential changes in tax rates over time and across verysimilar firms create an almost ideal experiment to study firm behaviorto taxation in a low enforcement-capacity setting.

One other interesting feature of the reform is that it was given aretroactive effect. The tax increase was announced on June 6, 2010, butit was made applicable from the beginning of the tax year i.e. from July1, 2009. Thus, by the time firms learnt the tax change 94% of the taxyear 2009 had already elapsed. Generally, behavioral responses totaxation conflate real and evasion margins and there is no satisfactoryway to separately identify the two. The retroactive applicability,

https://doi.org/10.1016/j.jpubeco.2017.11.003Received 21 October 2016; Received in revised form 19 September 2017; Accepted 14 November 2017

☆ I thank the Editor and three anonymous referees for helpful comments. This research has greatly benefited from discussions with Michael Best, Robin Burgess, Michael Devereux,Anders Jensen, Henrik Kleven, Camille Landais, Matthew Skellern, Johannes Spinnewijn and numerous seminar participants. Financial support from the International Growth Centre(IGC), Pakistan Program is gratefully acknowledged.

E-mail address: [email protected] According to a recent survey (Fuest and Riedel, 2009), informal sector on average constitutes one-third of the official economy in developing countries.2 The starting income tax rate, for example, in Pakistan during 2006–2009 was just 0.5%.3 Because the tax schedule is not indexed to inflation, such revision, involving the movement of bracket cutoffs, is needed every three to four years to avoid bracket creep.

Journal of Public Economics 157 (2018) 41–77

Available online 21 November 20170047-2727/ © 2017 Elsevier B.V. All rights reserved.

T

http://www.sciencedirect.com/science/journal/00472727

https://www.elsevier.com/locate/jpube

https://doi.org/10.1016/j.jpubeco.2017.11.003


mailto:[email protected]


http://crossmark.crossref.org/dialog/?doi=10.1016/j.jpubeco.2017.11.003&domain=pdf

however, allows me to disentangle tax evasion and real response in atransparent manner: while the post-2009 response to the reform couldencompass both margins, the 2009 response would comprise tax eva-sion mainly. This interpretation rests on the assumption that the reformwas not known before its official announcement. Tracking the entry oftreated and untreated firms over time, I provide a comprehensive testconfirming that the reform indeed was not anticipated.

I use administrative data from the Federal Board of Revenue,Pakistan (FBR), which comprise the population of income tax returnsfiled in 2006–11 and a set of firm characteristics. To guide the empiricalanalysis, I set up a simple model of firm behavior, characterizing therevenue and welfare implications of the tax change in terms of estim-able behavioral elasticities. The empirical strategy, motivated by thedifferential changes in tax rates across firms and over time, comparesthe evolution of partnership outcomes with that of sole proprietorshipand corporate outcomes in event-study research designs. The claim hereis not that a firm’s organizational form is randomly assigned; it is ratherthat the outcomes would have evolved similarly had the tax rates notchanged.

In the initial set of empirical results, I provide nonparametric evi-dence cataloging four important impacts produced by the reform. First,following the tax increase the number of formal partnerships declineddramatically: by 41% in 2009, by another 27% in 2010, and by anadditional 15% in 2011. This means that within three years of the taxincrease, the number of partnerships in Pakistan had declined to 36% ofthe baseline level. Second, partnerships which did not exit reportedconsiderably lower income: the average within-firm earnings growth,which consistently averaged around 8% in periods leading up to thereform, dropped by more than 50 percentage points in 2009. Third,there was significant income shifting towards the sole proprietorshipbusiness form: the number of partnership owners reporting positive soleearnings went up by 18% in 2009. Fourth, there was no discernibleincome shifting towards the corporate business form as only a fewpartnerships became corporations even after the tax disadvantage ofdoing so was largely removed. Using the research designs, I translatethese responses into behavioral elasticities and compute the welfarecost of the reform. The responses created by the reform are so large thatby the third year of its introduction the government was collecting lessrevenue than it would have without the tax increase. This implies thatthe new, flat tax rate of 25% was on the wrong side of the Laffer curveand would not have been optimal under any social preferences.

Exploiting retroactive applicability, I characterize the nature of theobserved responses. I argue that the predominant mechanism under-lying the intensive margin response – the tax-driven changes in re-ported earnings conditional on participation – was tax evasion. It isbecause the 2010–11 responses, which potentially conflate both realand evasion margins, were not different from the 2009 response, whichcaptures tax evasion mostly. I am, however, less certain whether theextensive margin response – the tax-driven changes in the number oftax-paying firms – captures firms exiting into informality or firmsshutting down completely. It is because the two extensive marginchoices I observe in the data – firms reporting zero earnings or dis-appearing completely after the reform – are potentially consistent withboth explanations. However, considering the structure of social in-surance in Pakistan, in particular that the owners of the exited firmswould not be eligible for any government assistance and would have towork to finance consumption, it is highly likely that the extensive re-sponse in large part reflects exit into informality.4

One key assumption underlying the sufficient statistics approachcommonly used for welfare analysis in the tax responsiveness literature(Feldstein, 1999; Chetty, 2009b; Saez et al., 2012) is that the tax change

does not generate significant externalities such as income shifting. Incontexts where this assumption is unreasonable, it is necessary to eitherestimate the consequences of the tax change on other bases directly orto adjust the welfare measure on the basis of some assumption on theseconsequences. This paper takes the former approach. The Pakistanicontext permits simultaneous identification of earnings responses andfiscal externalities arising out of the tax increase. I, therefore, estimateone negative – spillover effects on the value-added tax base – and twopositive – income shifting towards sole proprietorships and corpora-tions – externalities created by the reform separately, and incorporatethem into the welfare computations directly.

This paper contributes to three different strands of literature.First, it adds to the literature that estimates behavioral responses totaxation using administrative tax return data (see Saez et al., 2012for a recent survey). Most of the existing studies in this literaturefocus on only one margin of response. This paper represents perhapsthe first effort that identifies all important margins of firm responseto taxation together. Uncovering the anatomy of response, especiallyits decomposition into intensive and extensive margins, is particu-larly important in developing countries because policies to mitigatetax evasion and encourage formalization do not necessarily overlap(see Bruhn and McKenzie, 2014 for evidence on policies to encouragefirm formalization in developing economies). In addition, using ret-roactive applicability of the tax reform I am able to separate the realand reporting margins. Such separation is generally not feasibleunless special tax variation is available (Carrillo et al., 2017 andBachas and Soto, 2017 are two other recent studies that separate thereal and reporting responses). On the methodological standpoint,this study has the advantage that the tax variation created by thereform is not correlated with the prereform earnings, and conse-quently it does not face the principal identification challenge facedby other studies in this line of literature, that is mean reversion (seeSaez et al., 2012; Kopczuk, 2012 for a discussion on this issue).

Second, another important strand of literature estimates taxevasion and studies its relationship with the marginal tax rate(Andreoni et al., 1998; Slemrod and Yitzhaki, 2002). Due to the well-documented difficulties, only a handful of studies (such as Fismanet al., 2004; Marion and Muehlegger, 2008; Kleven et al., 2011; Bestet al., 2015; Waseem, 2017) are able to identify tax evasion cleanly.Even more difficult is to pin down its relationship with the tax rate:the comparative statics of evasion with respect to the marginal taxrate are highly sensitive to modeling assumptions (Slemrod andYitzhaki, 2002), and both the sign and magnitude of the effect areopen empirical questions (Kleven et al., 2011). This paper identifiestax-driven evasion cleanly and demonstrates that at least for risk-neutral agents in a low-enforcement setting it responds positively tothe marginal tax rate.

Finally, this paper is related to studies that examine the impact oftaxes on business organization choice of firms (see for example Gordonand MacKie-Mason, 1997; Goolsbee, 2004; Gordon and Slemrod, 2000).None of the existing studies, however, looks at the question from adeveloping country perspective, where returns to different businessforms could radically be different from those in rich countries.

The rest of this paper is organized as follows. Section 2 developsconceptual framework, Section 3 provides an overview of the contextand data, Section 4 describes the research design, Section 5 reports theempirical results, Section 6 computes the welfare costs of the reform,and Section 7 concludes.

2. Conceptual framework

This section develops a simple model of firm behavior under im-perfect enforcement to highlight the channels through which taxationaffects welfare in a developing country setting. The model captures keyelements of the tax environment, illustrating that increased taxationcan induce firms to (i) reduce output, (ii) increase tax evasion, (iii)

4 It is particularly true because I am able to show that migration within the formalsector – income shifting to sole proprietorships, corporations, and wage-earning sectors –is swamped by migration out of the formal partnership sector.

M. Waseem Journal of Public Economics 157 (2018) 41–77

42

change business form, or (iv) exit into informality. I first consider thedecision problem of a single firm and then extend the analysis to allowfor heterogeneity.

2.1. Setup

Consider a firm that decides how much output y to produce; howmuch earnings e to evade; and whether to operate as a sole proprie-torship, a partnership, a corporation, or an unorganized firm in theinformal sector. The production and evasion possibilities offered by thefour business forms are characterized by the production costs cj(y) andevasion costs gj(e), where j ∈{s,p,c,u} indexes the business form. Evasioncosts include, inter alia, the expected tax and fines that would be re-covered in case the evasion is detected.5 The firm faces a perfectlyelastic demand and can supply as much output as it desires at a fixedprice, which has been normalized to one. If the firm decides to operateas type j, it obtains the following after-tax profits

= − − −π y c y g e T y e( ) ( ) ( , ),j j j j j j j j j (1)

where Tj(.) is the tax liability faced by the firm given by

= − −T y e τ y μ c y e( , ) . [ ( ) ].j j j j j j j j (2)

The tax liability depends on tax rate τj and reported earnings zj ≡yj−μjc(yj)−ej, where μj ∈ [0,1) represents the fraction of the produc-tion costs that are allowed to be deducted from revenue for tax pur-poses. The parameter is allowed to vary across business forms becausethe tax rules governing the deduction of costs are slightly different forthe corporate and noncorporate forms. The firm makes its output andevasion choices by maximizing after-tax profits, producing the fol-lowing first-order conditions

=−

−′c y

τμ τ

( )1

1j jj

j j (3)

≥′g e τ( ) .j j j (4)

These conditions under the strict convexity of the two cost functionsimply that optimal output yj(τj) is a decreasing and optimal evasionej(τj) is a nondecreasing function of the corresponding tax rate τj.6

I assume that the firm has full information on the production andevasion costs so that it can compute after-tax profits from the fouroptions before choosing its business form. It, therefore, decides to op-erate as type j if the profits from doing so are at least as large as thosefrom the other options j′

≥ ∀ ≠′′

′ ′ ′ ′π y τ e τ π y τ e τ j j( ( ), ( )) max { ( ( ), ( ))} .j j j jj

j j j j (5)

Intuitively, the attractiveness of a business form to the firm is de-termined by its ability to produce and conceal earnings of the giventype. For example, it may decide to operate in the informal sector ifgu(e) ≈ 0 and cu(y) ≈ cj(y); ∀j≠u, which could be the case for a small,labor-intensive firm. On the other hand, it may decide to operate as acorporation if cc(y) ⋘ cj(y); ∀j≠c, which for example could be the caseif the firm stands to gain a lot from issuing capital in its own name. It isimportant to note that, in distinction to the output and evasion choices,

the firm’s choice of business form is a function of all tax rates andtherefore may change if any of the rates changes.7

2.2. Heterogeneity

The decision problem of a single firm carries over naturally to asetting with many heterogeneous firms. Each firm is characterized nowby a vector θ of firm-characteristics, which at the minimum includesthe two cost functions cj(y) and gj(e) that determine its ability to gen-erate and conceal income of type j.8 I assume that there are a continuumof firms of measure one that draw idiosyncratic θ from a smooth dis-tribution F(θ). These firms make their output, evasion, and businessform choices according to conditions (3)–(5). In heterogeneous setting,condition (5) implicitly defines the set of values for θ that induces firmsto choose type j conditional on the vector of tax rates τ chosen by thegovernment9

= ≥ ∀

≠

′′

′ ′ ′ ′τ θ θ θ θ θM π y τ e τ π y τ e τ j

j

( ) { : ( ( , ), ( , )) max { ( ( , ), ( , ))}

.

j j j j jj

j j j j

(6)

Thus, under the assumption that ties occur with zero probability, afraction ξj of firms operate as type j

∫=∈

τ θ θξ dF( , ) ( ).θ τ

jM ( )j (7)

The smooth distribution of θ among firms in this way translates intosmooth distributions of output and evasion within each set Mj(τ). Theconditions (3)–(5), therefore, should be viewed as a mapping that for agiven tax rate τ transforms F(θ) into four empirical distributions ofreported earnings F(zj(τ)).

2.3. Welfare

The principal interest of this paper is to investigate how taxationaffects firms’ output, compliance and business-form choices.Specifically, the empirical application considers a reform that increasesthe tax rate on partnership earnings (τp) and investigates its impacts onthe firm choices. To characterize the normative implications of thesechoices, I define social welfare simply as the sum of private surplus andpublic revenue

∫∑= +∈

τ θ θ θ θ θ θW π y τ e τ T y τ e τ dF( , ) { ( ( , ), ( , )) ( ( , ), ( , ))} ( ).θ τj M

j j j j j j j j( )j

(8)

Denoting aggregate partnership earnings by

∫≡ − −∈

τ θ θ θ θ θZ y τ μ c y e τ dF( , ) { ( , ) ( , ) ( , )} ( ),θ τ

pM

p p p p p p( )p (9)

the change in social welfare caused by a small increase in tax rate dτpcan be written as

5 I assume that the costs of evasion gj(e) are predominantly resource costs and not justtransfers across agents (see Chetty, 2009a for the distinction between the two). The as-sumption is motivated by the observation that in developing economies tax evasion istypically achieved at the cost of a loss in productivity. The productivity loss occurs fromactivities needed to hide real earnings from government, such as operating in cash.

6 Note that the right hand side of Eq. (3) is decreasing in τj as long as μj<1. Withμj=1, production costs are fully tax-deductible and the tax system does not distort outputdecision of the firm. The Pakistani tax code does not permit complete adjustment of costs.Specifically, the owners of sole proprietorships and partnerships are expressly barredfrom claiming wages from the firm. Similarly, corporate firms are not allowed to claim adeduction for the depreciation of tangible assets over and above the prescribed rates.Consistent with these provisions, I assume that μj is bounded from above by one.

7 A related point is that while the firm’s output and evasion choices depend on thecorresponding marginal tax rate, its business form choice depends on the average taxrates. In this model, I assume proportional taxation implying that the marginal andaverage tax rates are the same. In the Pakistani context, this assumption is not restrictive,as the tax system is proportional within brackets.

8 Note that the production and evasion costs differences across the organizationalforms primarily arise from the inherent features of the organizational form. For example,complementarity in skills of entrepreneurs makes partnership between them more pro-ductive than sole proprietorships. Similarly, evasion is more feasible in single-owner firmsthan in multi-owner ones. But such differences could also reflect entrepreneur-char-acteristics correlated with the choice of organizational form. For example, high-evasion-propensity entrepreneurs might be attracted towards sole proprietorships because of thegreater ease they offer to evade. In the model, the differences arising from both sourcesare captured in a reduced-form way through the parameter vector θ.

9 The treatment of firms’ discrete choice of choosing a business form here follows thestandard random-coefficients, discrete choice models (see for example Nevo, 2000).


43

⏟⏟ = −

−+ + +dW

dττ

τε η η η Z

1[ ] .

p

p

pp p sp cp p

Intensive Margin Extensive MarginIncome Shifting (10)

The above expression exploits the envelope condition, noting thatthe tax change has no first-order impact on private welfare because it isalready at the optimum. In addition, the direct effect of the tax increaseon firm profits cancels out, as the welfare criterion implicitly assumesthat the additional revenue is returned to firms in a lump sum fashion.This effectively reduces the welfare costs of the reform to its impact onpublic revenue only, which can be broken down into the followingthree terms.

2.3.1. Intensive margin responseThis term reflects that partnerships might reduce output and/or

increase evasion following the tax rate rise, according to conditions (3)and (4). These two effects act in the same direction, reducing govern-ment revenue from the tax base. The aggregate effect is proportional tothe elasticity of reported partnership earnings εp and can be brokendown further into the two underlying margins

⏟= + −ε σ ε σ ε(1 ) ,p p p

yp p

e

Intensive MarginReal Response Evasion Response (11)

where εpy is the elasticity of real partnership base (the first two terms

inside the integral in Eq. (9)); εpe is the elasticity of tax evasion by

partnerships; and σp is the ratio of real and reported partnership base.These elasticities are aggregate elasticities defined with respect to thenet-of-tax rate 1−τp, and given that firms in this model are hetero-geneous are income-weighted averages of the corresponding firm-levelelasticities.

2.3.2. Extensive margin responseThis term captures firms that exit the partnership base after the rate τp

goes up. For these firms, the net value from operating as partnership j=pwas larger than the other options j′∈{s,c,u} at the prereform rates but not atthe post-reform rates. Some of these firms would change their business formto sole proprietorships or corporations, while the rest would disappear intothe informal sector. The size of the effect depends on the aggregate ex-

tensive margin elasticity defined as ≡ − ∂

∂ −η τ

τp

τξ

ξ

τ(1 )

( )

( )

(1 )p

p

p

p, which as earlier is an

income-weighted average of the firm-level elasticities.

2.3.3. Income shiftingThis term captures partnerships that change their business form

after the reform. Such income shifting offsets revenue loss from theextensive response mentioned above. The size of the offset dependsupon the two income shifting elasticities ≡ ∈− ∂

∂ −η k s c; { , }ττ

kpτ

ξξ

τ(1 )

( )( )

(1 )p

k

kp

,

where a different notation for the elasticities ( η ) is to emphasize thatthey are aggregated using revenue rather than income weights. Therevenue weighting here reflects that the shifted earnings face poten-tially different tax rate in the new tax base. In the extreme case whereall partnerships that exit reappear as sole proprietorships or corpora-tions contributing the same revenue as earlier, the income-weightedelasticity ηp and the sum of the two revenue-weighted elasticities

+η ηsp cp will cancel each other and there will be no change in thegovernment revenue or social welfare.

2.4. Spillover effect on the VAT base

The analysis so far has ignored one important feature of the taxenvironment that a subset of firms also remit VAT on their sales. Theincome-tax-driven changes in firm behavior would impact governmentrevenues from the VAT base as well, increasing the costs of the reformabove those given by formula (10). In Appendix A.2, I show how I in-corporate this fiscal externality into the welfare computations.

Three features of the framework above are idiosyncratic to the

Pakistani setting and need to be emphasized. First, I do not explicitlymodel a firm’s choice to shut down completely. This is because the datadoes not allow me to distinguish between firms producing zero outputand firms operating in the informal sector. The margin exit into in-formality in this paper therefore includes real exit.10 Second, the wel-fare analysis here focuses solely on the revenue effects of the tax re-form, ignoring the impacts on welfare operating through the input andoutput markets. For example, firms that leave the formal sector mightlay off workers, creating additional welfare losses that are not capturedin formula (10). I do not observe firms’ interactions in these marketsand therefore cannot take these into account. And finally, the frame-work I use is static in nature and abstracts from dynamic decisions suchas investment. It is an appropriate framework for the Pakistani settingbecause responses of relatively small, less capital-intensive firms over ashorter horizon of up to three years are considered.

3. Institutional background and data

This section describes institutional features of the Pakistani setting,focusing in particular on changes in the tax treatment of corporate andnoncorporate firms in the country between 2006 and 2011.

3.1. Taxation of firm profits in Pakistan

Consistent with the international practice, Pakistan has two sepa-rate regimes for the taxation of corporate and noncorporate firms.Profits of noncorporate firms – sole proprietorships and partnerships –are taxed through the personal income tax schedule. In periods prior tothe reform, a single tax schedule was applicable to earnings of bothtypes of firms. It consisted of fourteen brackets with a fixed average taxrate, varying progressively from 0% at the bottom to 25% at the top,assigned to each bracket. The reform, announced on June 6, 2010, re-placed this schedule with two different tax systems. For partnerships, anew flat-tax scheme involving a tax rate of 25% with no exemptionthreshold was introduced. The change was applied retroactively fromJuly 1, 2009, so that partnership earnings corresponding to tax year2009 and onward were subject to the new tax rate. For sole proprie-torships, the progressive tax schedule was maintained, but the numberof brackets was reduced from fourteen to six and the bracket cutoffswere moved. The new schedule was applied prospectively from July 1,2010, so that sole-proprietorship earnings corresponding to tax year2010 and onward were subject to the new tax rates. The new schedulegenerally maintained the prereform rates but the movement of thebracket boundaries meant that sole proprietorships in some areas of theincome distribution experienced a slight reduction in tax rates.

In contrast to noncorporate firms, profits of corporate firms inPakistan were always taxed at a flat rate of 35%. Small companiesdefined as corporations which (i) register after June 2005, (ii) have nomore than 250 employees, (iii) have annual sales up to PKR 250 mil-lion,11 (iv) have paid-up capital up to PKR 25 million, and (v) have notbeen formed by the splitting up or reconstitution of a company alreadyin existence were allowed a concessionary tax rate of 20%. Such smallcompanies comprise less than 15% of the corporate sample. During theperiod 2006–2011, the standard tax rate on corporate earnings stayedunchanged at 35%, but the rate applicable to small companies wasincreased to 25% from 2010.

10 Conflating these two margins, however, is not as restrictive as it seems, especially ifwe take into account the structure of social insurance in Pakistan. The country has only asmall means-tested income transfer program, targeted to extreme poor. Since income taxexemption threshold is set relatively high, the owners of firms dropping out of the formalsector would not be eligible for any government support and would have to work togenerate consumption. Thus, as long as the costs of operating in the formal sector are nottoo high, it is natural to expect that firms dropping out of the formal sector would chooseto operate informally rather than shut down.

11 The PKR-US$ exchange rate hovered between 60 and 90 during 2006-11.


44

Fig. 1 illustrates the tax variation created by the reform: the toppanel compares the pre- and post-reform tax rates, and the bottompanel plots the evolution of average tax liability experienced by thethree types of firms.12 As a result of the reform, the average tax liabilityfaced by partnerships firms quintupled, increasing from 5% to 25% ofearnings in 2009. In contrast, the average tax liability faced by soleproprietorships remained unchanged up to 2009 but decreased slightlyfrom 2010 when the revisions to the existing tax schedule became op-erational. The average corporate tax liability stayed almost the samethroughout the sample period.

3.2. Registration and filing rules

All firms with earnings above the exemption cutoff are required toregister with Pakistan’s tax authority, the Federal Board of Revenue(FBR). On registration, partnerships and corporations are assigned aunique tax identifier. Sole proprietorships, on the other hand, areconsidered indistinguishable from their owners: the firm and ownershare the identifier and file a common tax return. As long as they areregistered, firms are required to continue filing returns even if theirincome falls below the exemption cutoff.

A firm can change its business organization at any time. If a part-nership decides to become a corporation, it needs to get itself registeredas a company with the Security and Exchange Commission of Pakistan(SECP). Before such incorporation, the firm has to re-register with theFBR as a company and to get a new identifier. Incorporation is a costlyprocess, as in addition to paying a fee, which begins from PKR 5000 andincreases with the issued capital, the firm is required to register with ahost of other departments and regulatory institutions. In contrast, if apartnership breaks up and the owners desire to continue the dividedbusiness as sole proprietors, no regulatory approval is needed. Theowners can do it on their own, reporting earnings of the new soleproprietorships in their personal tax returns. These rules have im-portant implications for identifying income shifting from partnershipsto the other business forms. Specifically, income shifting to corporatefirms, if it happens, would leave two markers: (i) the entry of newcompanies would increase because of the fresh registration require-ment, and (ii) former partnership owners would start reporting positivedividend income in their personal returns. Compared to this, incomeshifting to sole proprietorships would manifest itself only in the per-sonal returns of the former partnership owners.

3.3. Data

I use administrative data from the FBR that include the universe ofincome tax returns filed in 2006–2011 and a set of taxpayer char-acteristics reported at the time of registration. The tax return datasetcontains variables corresponding to line items reported on the returnform, including a brief profit and loss account, the decomposition oftaxable income by source, and tax computations. The registration da-taset includes individual and firm characteristics, such as date of re-gistration, industry, and region. Since July 2009, electronic return filingis mandatory for all firms other than small sole proprietorships.Consequently, most of the 2008–2011 returns used in this study havebeen filed electronically.13 The rest of the returns were filed at desig-nated bank branches and were fed into computers by an IT firm distinctfrom the FBR. Throughout the period covered by this study, the FBR hasbeen using the data for automated processing and payment of VAT andincome tax refunds, which has ensured that the data were kept updatedand relatively free from errors.14

Table 1 presents the summary statistics of the data for the baselineyear 2008. All empirical results in this paper, unless otherwise stated,are based on the analysis sample, which contains firms that have baseperiod income (zit) in the range [0 650,000]. The analysis samplecontains around 95% of all partnerships in the sample (row 2 of thetable). I exclude firms in the rest of the income distribution becausethey experience relatively smaller tax changes and the density of taxfilers in the region is too thin to estimate responses credibly.

Expectedly, annual sales and earnings of partnerships (rows 1 and 4)are on average lower than those of corporations and higher than those

A: Tax Rates0

10

20

30

40

Ave

ra

ge

Ta

x R

ate

(%

)

0 200 400 600 800 1000 1200 1400

Taxable Income in PKR 000s

Partnerships (2009−11)

Sole Proprietorships (2006−09) & Partnerships (2006−08)

Sole Proprietorships1

(2010−11)

Corporations2

(2006−2011)

B: Average Tax Liability

010

20

30

40

Ave

ra

ge

Ta

x L

iab

ility (

% o

f T

axa

ble

In

co

me

)

2006 2007 2008 2009 2010 2011

Partnerships Sole Proprietorships Corporations

Fig. 1. Tax variation created by the reform. 1. Exemption threshold for sole proprie-torships in 2011 was PKR 350,000. 2. For small corporations the tax rate was 20% whichincreased to 25% from 2010 (see Section 3.1 for details). Notes: The figure displays taxvariation created by the reform. Panel A plots the tax rates applicable to the three types offirms from 2006 to 2011. The Pakistani tax code prescribes average rather than marginaltax rate in a given bracket of income, and all curves accordingly show the average tax rateas a function of annual taxable income. Taxable income is shown in thousands of Pa-kistani Rupees (PKR). The PKR-USD exchange rate was about 60 in 2006 and increased toaround 90 in 2011. Panel B plots the evolution of average tax liability experienced by thethree types of firms from 2006 to 2011. The average has been estimated on the actualsample of filers in each year and has been defined as the aggregate tax liability as apercentage of the aggregate taxable earning of the type of firms in the year.

12 The average tax liability has been estimated on the actual sample of filers in eachyear and has been defined as the aggregate tax liability as a proportion of the aggregatetaxable earning of the type of firms in the year.

13 Returns for a year t are filed in the September of year t+1. The electronic-filingprovision, therefore, applies to all 2008–2011 tax year returns.

14 It is important to emphasize that the Pakistani tax system is based on the principal ofself-assessment, meaning thereby that all filed returns are considered final unless selectedfor audit. Each year, the tax administrations audits a small sample of returns. I, however,do not observe the incidence or outcome of the audits.


45

of sole proprietorships. Similarly, in terms of other characteristics thatdetermine a firm’s propensity to comply with tax laws (rows 6–11)partnerships lie in between the other two business forms. This is par-ticularly helpful, as having a control group on either side of the com-pliance scale acts as a natural robustness check on the internal validityof the difference-in-differences estimates. Appendix Fig. A1 comparesthe industry, geographic, and size distribution of firms. While the in-dustry and size distribution of firms is fairly similar, the geographicdistribution is not: corporations are mostly located in the top three ci-ties of Pakistan, whereas partnerships and sole proprietorships aredistributed symmetrically throughout the country. To account for thegeographic disparity, I also report results from specifications that in-clude the region fixed effects.

3.4. Was the reform anticipated ?

I use retroactive application of the tax increase to characterize thenature of the observed responses (real response vs. tax evasion). Suchcharacterization requires that the reform was not known before its officialannouncement in June 2010. To assess this, Fig. 2 compares the entry ofnew partnerships and corporate firms in Pakistan. For this particularquestion looking at the entry is helpful as the data are available at a dailyfrequency and given the large size of the tax shock the time around whichthe tax increase became known can be identified from a break in trend.Panel A plots the raw data, aggregating the entry to a monthly frequency.Panel B plots the coefficient and 95% confidence interval from a difference-in-differences regression on the two entry series in Panel A. Of the 47preannouncement months considered here (July 2006 to May 2010), theDD coefficient is statistically insignificant in 33 months, including themonth immediately preceding the announcement. By contrast, the coeffi-cient is negative and statistically significant in all 34 post-announcementmonths. The plots thus show that partnership and corporate entry were onreasonably parallel trends in periods leading up to the reform and that itwas only after June 2010 that the partnership entry began deviating fromthe corporate entry systematically. Appendix Fig. A2 repeats the analysis

using sole proprietorships as control, reinforcing the conclusion that thereform was not anticipated.15 This, in fact, should not be surprising as thePakistani authorities are very secretive about tax changes, fearing thattaxpayers might shift activity across time or entities to minimize their taxpayments.

3.5. Income shifting costs

Eq. (5) shows that a firm’s business form choice is a function of its abilityto produce and conceal earnings of a given type, captured by the two costfunction – cj(.) and gj(.) – and the three marginal tax rates τj; j∈{s,p,c}. Giventhat in the Pakistani setting the three marginal tax rates are in general notequal, a firm’s choice of reported earnings and organizational form can beused to infer its productivity in generating reported earnings of the giventype. To see this, consider a partnership that produces y units of output andevades e units of income, facing a tax liability of τp [y−μpcp(y)−e]. Usingtax rules and data from returns filed by its owners, I am able to compute thefirm’s counterfactual tax liability had it operated as a sole proprietorshipchoosing the same level of output and evasion. Under the assumption thatthe parameter μj does not vary across the two business forms, the differencerepresents the minimum earnings boost that the firm experiences fromoperating as a partnership. Since the firm would lose at least this much ofprofits if it decides to become a sole proprietorship, the difference providesa lower bound on income shifting costs in the neighborhood of the op-timum.

Fig. A3 plots the distribution of these costs for partnerships in2006–11. The histograms show that these costs are generally quitelarge, suggesting that the change of organizational form is not a trivial

Table 1Summary statistics.

Full sample Analysis sample

Partnerships Sole props. Corporations Partnerships Sole props. Corporations(1) (2) (3) (4) (5) (6)

Outcomes1. Taxable Income 442,114 350,234 83,428,856 223,319 147,171 200,958

(1,829,383) (20,390,724) (1,173,785,216) (124,759) (74,017) (175,916)[198,100] [125,000] [740,074] [185,000] [125,000] [147,407]

2. Number of FirmsTaxable Income>0 21,319 373,279 5122 19,357 365,686 2461Taxable Income = 0 16,551 118,573 12,597 16,551 118,573 12,597Characteristics3. Annual Sales 32,512,930 4,768,342 806,324,032 14,791,702 3,838,501 50,771,6044. Tax Liability 68,307 55,154 29,173,054 13,149 4436 63,9385. Age 4.08 7.29 6.75 3.95 7.25 5.756. Electronic Filer 0.56 0.05 0.99 0.55 0.05 0.987. VAT Registered 0.24 0.06 0.52 0.21 0.06 0.378. Round Filer 0.41 0.67 0.02 0.44 0.68 0.039. Buncher 0.28 0.36 – 0.31 0.37 –10. Dominated 0.03 0.02 – 0.03 0.02 –11. Revised Return 0.01 0.00 0.05 0.01 0.00 0.0412. Large City 0.39 0.37 0.76 0.37 0.37 0.7113. Withholding Agent 0.14 0.01 0.92 0.09 0.01 0.90

Notes: This table presents descriptive statistics of the data for the baseline year 2008. Analysis sample contains firms with earnings in the range [0 650 K], whereas the full samplecontains all firms with nonnegative earnings. The first row compares the mean, standard deviation, and median of taxable income reported by the three types of firms. The standarddeviation and median are in parenthesis and square brackets respectively. Rows 3–13 compare the mean of the firm characteristic variable across the three types of firms. Annual salesand tax liability are reported in PKRs. Age is defined as the number of years a firm has been registered with the FBR. Round filer is defined as a firm which reports earnings in exactmultiples of thousands. It is generally considered a good indicator of the quality of record keeping within the firm. Buncher and Dominated are the sole proprietorships and partnershipswith earnings within the bunching and strictly dominated regions around the notches in the baseline tax system (2006–08) as defined in Kleven and Waseem (2013) . Revised returnindicates that a firm files a revised return for 2008 to rectify any mistakes in the original return. Large City indicates that a firm has its head office in one of the three big cities of Pakistan— Lahore, Karachi, and Islamabad. The detailed description of the rest of the variables is provided in Appendix A.1.

15 For sole proprietorships, however, I observe the date of registration only if the firmfiles a return, as they are not required to register separately from their owners. Theanalysis in the figure, accordingly, is limited to the subset of firms which file tax return inthe sample period. For this reason, the two entry series in the figure decline mechanicallyover time and are more noisy. Nevertheless, the results are consistent with those fromcorporate firms as control.


46

decision for a firm and involves a real loss in productivity.16 Note that Icannot conduct similar analysis for partnerships vs. corporations, as thecost deduction rules (μj) are different for the two business forms. Thevariation in μj means that I am unable to compute the counterfactual taxliability of a partnership if it operates as a corporation from the avail-able data.

4. Research design

This section describes the difference-in-differences research designs

I use to estimate the parameters in formula (10). The idea behind theresearch designs is to compare partnership outcomes with sole pro-prietorship and corporate outcomes over time to isolate the tax-driveneffects. The claim here is not that a firm’s business organization israndomly assigned; it is rather that the outcomes would have evolvedsimilarly had the tax rates not changed.

4.1. Intensive margin

To obtain the intensive margin elasticity, I estimate the followingmodel

= + + − + + +X δz α β Partnership ε τ λ uΔ log Δ log(1 ) ,it i it i t it

(12)

where i and t index firms and tax years; Δlog zit is within-firm log change inreported earnings from period t−1 to t; Partnershipi is a dummy indicatingthat i is a partnership, Δlog(1−τit) is within-firm log change in net-of-taxrate from period t−1 to t; Xi are a set of controls; and λt are year fixedeffects. To address the potential endogeneity of tax rate to the choice ofreported earnings, I use tax variation created by the policy reform only andinstrument Δlog(1−τit) in the first-stage with the double-difference inter-action term Partnership×Postit, where Postt is a post-reform period indicator.I estimate the model using sole-proprietorships and corporations as controls,reporting the results from the alternative specifications separately. Thebaseline specification does not include any controls, but I show that theresults are robust to including industry, region, and tax office fixed effects.17

All estimates are weighted by income so that the elasticity estimate corre-sponds to the parameter εp in formula (10).

There are three potential threats to identification in this setting. First,reported earnings might not be on a common trend in the treatment andcontrol groups. Second, the composition of the sample might change in thepost-reform periods in a way creating a correlation between the double-interaction and error terms. Third, the control outcomes might also be af-fected by changes introduced by the reform. I take the following precautionsand/or conduct robustness checks to rule out these concerns.

I present three pieces of evidence supporting the common trends as-sumption. First, I show visually that the preexisting earnings trends wereparallel in the treatment and two control groups. In fact, the trends were sostable and flat that even the time-series estimates are credible in this setting.Second, I supplement all regression-based estimates with placebo analysis,pretending that the reform took place one year earlier than it actually did.Third, I illustrate that the result are not affected when the year fixed effectsare replaced by linear, separate linear, and industry-specific time trends.

I also estimate Eq. (12) on balanced-panel samples. These samplesinclude only firms which report positive earnings in all years consideredin the analysis. They, thus, do not allow entry and exit, holding thecomposition of the treatment and control samples fixed for the entireperiod of estimation. The results from the balanced-panel samples arealways comparable to those from the complete samples, suggesting that(i) concerns from a change in the composition of the sample caused byendogenous entry and/or exit of firms are not important in this setting,and (ii) the intensive margin responsiveness estimated from Eq. (12) isbroadly representative of the responsiveness in the population.18

Finally, I take two precautions to ensure that the control outcomesare not contaminated directly or indirectly by the tax changes. First, to

A: Entry of New Firms0

50

01

00

01

50

02

00

02

50

0

Nu

mb

er o

f N

ew

En

tra

nts

2006 2007 2008 2009 2010 2011 2012

Partnerships Corporations

B: Difference-in-Differences

−2000

−1000

01

00

02

00

0

Diffe

re

nce

−in

−D

iffe

re

nce

s C

oe

ffic

ien

t

2006 2007 2008 2009 2010 2011 2012

Coefficient 95% Confidence Interval

Fig. 2. Was the reform anticipated? Notes: The figure investigates if the reform wasanticipated before its official announcement on June 6, 2010. Panel A of the figurecompares the entry of partnerships with that of corporations from July 2006 to March2013. Each dot on the two curves represents the number of firms that get registered withthe tax authority in that particular calendar month. Year t on the x-axis indicates the firstmonth (July) of the tax year t. Dashed vertical line demarcates June 2010. Panel B of thefigure plots the coefficients from a difference-in-differences regression on the two series inPanel A. Each dot on the solid curve shows the coefficient for the particular month, in-dicating the additional entry of partnerships in the month relative to corporations. Thegray area plot shows the 95% confidence interval around the coefficient.

16 Theoretically, the income shifting costs could reflect either that firm operating aspartnerships are more efficient in producing output cp(y)< cs(y) or that they are able tohide income more easily gp(e)< gs(e) relative to if they operate as a sole proprietorships.While I am unable to break down the costs into these two components, the fact thatpartnerships have attributes that are generally negatively associated with tax evasion –for example, on average they are larger, have higher earnings, have greater fraction oftheir earnings reported by third parties, are more likely to be electronic-filers, and re-spond less aggressively to tax incentives (Table A20) – suggests that the income shiftingcosts reflect in large part the ability to produce output more efficiently.

17 Pakistan has two types of tax offices: three Large Taxpayer Units (LTUs) and twelveRegional Tax Offices (RTOs). Including tax office fixed effects accounts for the possibilitythat firms administered by different offices might have been exposed to varying levels ofenforcement.

18 It is important to emphasize that firms which exit in 2009 do not feature in either ofthe two samples on which Eq. (12) is estimated. While the comparability of the elasticityover time and across samples provides strong evidence that these firms would have re-sponded similarly to the other firms had they remained active, the intensive margin re-sponses of these firm are not directly identifiable. If these firms were different from theother in terms of their intensive-margin responsiveness, the estimates from Eq. (12)would represent the responsiveness among the active firms only and not the population.


47

eliminate indirect effects operating through income shifting, I dropfirms whose owners hold an interest in a partnership in any period fromthe control groups. This, however, turns out to be a careful precautiononly, as such firms constitute less than 5% of the control sample and theresults with or without them are indistinguishable. Second, to addressthe concern that a subset of control firms are affected directly by taxchanges from 2010, I first estimate Eq. (12) on a period 2006–09 pro-ducing a short-run estimate of the elasticity. I then re-estimate Eq. (12)on a period 2006–2011, excluding the subset of control firms affectedby the tax changes. Such exclusion is immaterial for the corporatecontrol group, as the relevant tax change was (i) applicable to onlyaround 15% of firms; (ii) a function of predetermined firm-character-istics; (iii) a tax increase, meaning that even if we ignore it the resultingbias would only push the estimates downwards. In contrast, the ex-clusion could be consequential for the sole proprietorship controlgroup, as the relevant tax change was a nonlinear function of the baseperiod income. The concern, however, turns out to be of little practicalrelevance, as the 2010–11 estimates from the two alternative controlgroups are either statistically indistinguishable from zero or econom-ically insignificant relative the 2009 estimate.19

4.2. Extensive margin

I use a three-step strategy to estimate the extensive margin elasti-city. The first step in the strategy is to ascertain the counterfactualnumber of tax filers in the post-reform periods. For this purpose, I es-timate a difference-in-differences model similar to Eq. (12) using thelog number of filers as the outcome variable, where I consider a firm asa filer in period t if it reports positive taxable income in the period.20 Onthe basis of the fitted model, I predict the counterfactual number ofpartnership tax-filer in 2009–11. The difference between the predictedand actual number of tax filers represents the extensive-margin re-sponse to the reform, which can be used to compute the correspondingunweighted elasticity.

To obtain the income-weighted elasticity needed for formula (10)and to explore response heterogeneity, I go a step further and estimatethe complete counterfactual distribution of partnerships earnings. Thisstep is predicated on the observation that for a given tax system theempirical earnings distribution does not change from year to year

≡ < = ∀F z τ z z τ F z τ t( | ) Pr ( | ) ( | ) .t p p it p p, (13)

This essentially implies that the macro-driven entry and exit of firms arenot correlated with firm-earnings, so that even if the number of tax filerschanges from year to year the density of tax filers stays the same. Ipresent a nonparametric test of this assumption in Appendix Fig. A4.The figure plots the observed Ft(zp) for the three preintervention periods2006–2008, showing that consistent with Eq. (13) the empirical CDF isindistinguishable across periods of no tax change. Given this statio-narity of the CDF, the counterfactual distribution can be constructed byadjusting the baseline empirical distribution to have the same mass aspredicted by the difference-in-differences model in the first step.

In the final step, I compare the observed and counterfactual dis-tributions to compute the income-weighted elasticity. Before makingthis comparison, I strip the observed distribution of the intensive-margin responses. Intuitively, this step is necessary because firms report

lower earnings after the tax rate increase, creating a leftwards shift ofthe observed distribution. Partialling out the intensive response ensuresthat any difference between the observed and counterfactual distribu-tion in a given area of the distribution identifies the tax-driven reduc-tion in the number of tax filers in the area. By weighting these localextensive response estimates with the taxable income in the area, Iobtain the aggregate income-weighted elasticity ηp.

I support the elasticity estimates from the strategy with the resultsfrom the following auxiliary regression

> = + + × + +

+

γ X δz α β Partnership λ

u

Φ Partnership Post1( 0) (

),it i it i t

it (14)

where Partnership × Post it is a vector of three interaction terms oneeach for 2009 to 2011, and all other variables are defined similarly to asin Eq. (12). I fit the equation using both probit and linear models onsamples containing for period t all firms that file return for the period,reporting earnings in the range [0 650 K]. The coefficients on the threeinteraction dummies reflect how the probability to report positivetaxable income changes for partnerships in the corresponding periodrelative to the control firms. Though the results from this exercise arequantitatively not comparable to those from the strategy above,21 itpermits conducting the robustness checks mentioned in the last sectionin a transparent, regression-based framework.

4.3. Income shifting

The registration and filing rules described in Section 3.2 imply thatif a partnership becomes a corporation, its owners would begin re-porting dividend income in their personal returns. Similarly, if a part-nership breaks up into sole proprietorships, its owners would beginreporting sole-proprietorship income in their personal returns. I, ac-cordingly, use the following model to identify income shifting frompartnerships to the other business forms

> = + + − + + +X δz α β Partner η τ λ uΦ1( 0) ( log(1 ) ),j it i it i t it,

(15)

where, 1(zj,it>0) is an indicator that i reports positive earnings fromsource j in period t, Partneri is a dummy showing that i was a partner-ship owner in any of the three prereform periods, log(1−τit) is the lognet-of-tax rate experienced by i in period t, and the rest of the variableshave the usual interpretation. To capture the incentive for incomeshifting, I simulate τit for treated taxpayers as the marginal tax rate thatwould apply if their source j income was reported as partnership in-come (it is flat 25% in the post-reform periods). For control taxpayers,the tax rate variable is computed as in the previous sections.22 Theestimates are revenue-weighted so that the elasticities from the equa-tion correspond to the two parameters, ηsp and ηcp, in formula (10). Iconduct the tests mentioned in the last two sections to establish therobustness of the results.23

The above model considers former partnership owners receivingpositive dividend income in the post-reform periods as an evidence of

19 An alternative strategy to estimate 2010–11 responses would have been to controlfor the tax changes experienced by control firms directly in the regression. This approach,however, requires that the elasticities for the control and treated firms are the same. Idecide against using this approach because the evidence in Waseem (2017) shows that theelasticity generated by tax reforms that reduce the tax rate to zero is uncharacteristicallylarge. Many sole proprietorships experience such a tax change in 2010.

20 To ensure that the number of tax filers in the control group are not affected throughincome shifting, I apply the safeguard mentioned in the last section here as well, droppingfirms from the control groups whose owners have been partners in a partnership firm inthe prereform periods. Again, it turns out to be a careful precaution only as the number ofsuch firms is negligible relative to the size of the control group.

21 The extensive margin response occurs through three distinct channels: (1) reducedentry; (2) increased exit comprising firms that exit and stops filing returns; (3) increasedexit comprising firm that continue filing returns but report zero earnings. While the es-timates from the strategy encompass all three channels, the estimates from Eq. (14) reflectthe last channel only.

22 Throughout this paper I compute marginal tax rate as ≡ + −τitT zit T zit( Δ) ( )

Δ, where

T(.) is tax liability and Δ represents a small increase in income (PKR 50).23 For consistency, I estimate Eq. (15) using both sole proprietorships and corporations

as control groups. Corporate firms, however, cannot have sole-proprietorship income.Therefore, specifications that estimate income shifting to sole proprietorships using cor-porate firms as controls compare the propensity to report positive sole-proprietorshipincome by former partnership owners to the propensity to report positive taxable incomeby corporations, attributing any difference in the post-reform years to the tax increase. Ipresent placebo estimates to establish that such comparison indeed captures the desiredtax-driven impact.


48

income shifting to corporations. The problem with this approach is thatcorporations may not issue dividends every year, so the model mayunderestimate the response. I, therefore, supplement the exercise with anonparametric permutation test, looking at the entry of new

corporations. Since partnerships that become corporations are legallyobliged to re-register, any impact of the reform on this margin caneasily be detected by tracking the registration of new corporations overtime. I test this using the following regression

A: Partnerships – Prereform B: Partnerships – Post-reform

Δm07 = 8.8%

Δm08 = 27.8%

0400

800

1200

1600

Num

ber o

f F

ilers

0 100 200 300 400 500 600


2006 2007 2008

Δm09 = −41.2%

Δm10 = −27.3%

Δm11 = −15.0%

0400

800

1200

1600

Num

ber o

f F

ilers

0 100 200 300 400 500 600


2008 2009 2010 2011

C: Sole Proprietorships – Prereform D: Sole Proprietorships – Post-reform

Δm07 = −6.2%

Δm08 = −4.4%

020

40

60

80

Num

ber o

f F

ilers 0

00s

0 100 200 300 400 500 600


2006 2007 2008

Δm09 = −5.9%

Δm10 = −2.4%

Δm11 = −13.6%

020

40

60

80

Num

ber o

f F

ilers 0

00s

0 100 200 300 400 500 600


2008 2009 2010 2011

E: Corporations – Prereform F: Corporations – Post-reform

Δm07 = 31.8%

Δm08 = 39.8%

03

06

09

01

20

15

0

Num

ber o

f F

ilers

0 100 200 300 400 500 600


2006 2007 2008

Δm09 = −6.7%

Δm10 = −10.5%

Δm11 = 1.4%

03

06

09

01

20

15

0

Num

ber o

f F

ilers

0 100 200 300 400 500 600


2008 2009 2010 2011

Fig. 3. Taxable income distribution. Notes: The figure compares the pre- and post-reform taxable income distributions across the three types of firms. Each dot on the curves representsthe upper bound of a PKR 10,000 bin and denotes the number of firms which report earnings within that bin. The notches in the 2006–08 schedule are shown by the vertical dotted lines(Panels A–D only). In the right-hand side panels, the 2008 distribution is shown again for comparison purposes. Yearly changes in the number of tax filers are shown by Δmt, which foryear t signifies the change in the number of filers from year t−1 to t as a percentage of the number of filers in year t−1.


49

= + + + × +Entry α β t γ Post δ t Post ν ,t t t t (16)

where Entryt refers to the number of new corporations registered inperiod t. This equation fits a linear trend on the pre- and post-reformentry series and tests whether this trend changes at the time of thereform. In addition to checking for the significance of δ , I compare it tosimilar coefficient in the placebo regressions estimated on the prere-form periods only.24 As I observe the entry of new corporations for alarge number of prereform periods, I am able to generate a completedistribution of the placebo coefficient. If the reform had a significantpositive impact on the entry of new corporations, the estimated δwould lie in the upper tail of this distribution.

5. Empirical results

In this section, I first present nonparametric evidence on how thenumber of and earnings reported by the treated firms responded to theincrease in the tax rate. Later, using the research designs detailed aboveI translate the responses into the behavioral parameters of interest.

5.1. Nonparametric evidence

Fig. 3A–B shows the distribution of earnings reported by partner-ships over the period 2006–11. The prereform plots (Panel A) illustratetwo key points. First, the number of partnerships was increasing beforethe reform: it increased by 9% in 2007 and by 28% in 2008. Second,despite the increase in numbers the shape of the distribution was re-markably stable and did not change from one year to the other. Thisdemonstrates that the entry and exit during the periods of no tax changeare not correlated with firm-earnings, providing a direct evidence insupport of the assumption Eq. (13). Fig. 3B plots the 2008 and the threepost-reform distributions together, depicting the enormous impactproduced by the tax increase. Not only was the increasing prereformtrend reversed, but also the number of partnerships started decreasingsharply after the reform. The number decreased by 41% in 2009, byanother 27% in 2010, and by a further 15% in 2011. This means thatwithin three years of the tax increase the treated tax base had shrunk to36% of the baseline level.

In addition to the large extensive-margin response, the plots alsocarry the signature of the intensive margin response: The post-reformdensities are higher relative to the prereform densities at the bottom ofthe distribution (earnings < 100,000). It shows that partnershipswhich did not exit reported lower earnings after the reform, creating aleftwards shift of the empirical distribution.

To demonstrate that the observed responses are driven by the taxincrease and not by any macroeconomic shocks, I present in Fig. 3C–Fthe corresponding distributions of sole proprietorship and corporateearnings. In constructing Panels C–D, I (i) drop sole proprietors thatreport any income from a partnership in 2006–11 and (ii) strip the2010–11 distributions of intensive responses to the 2010 tax changesusing the assumption Eq. (13). The control group earnings distributionsin 2006–11 (Panels C–F) are almost on top of each other, showing nodiscernible change in outcomes over time. This confirms that the large-scale erosion of partnership earnings depicted in Panel B was caused bythe tax increase. Appendix Fig. A5 repeats the analysis without makingchanges (i) and (ii) to the sole-proprietorships distributions, illustratingthat the changes do not make any material difference to the conclusion.

5.2. Elasticity estimates

5.2.1. Intensive marginGraphical evidence — Fig. 4 compares the evolution of reported

earnings across the treated and control firms in the period 2006–11.The figure is based on the analysis sample, containing firms with po-sitive earnings only i.e. zit ∈ (0 650 K]. It thus isolates the earningsresponse conditional on participation produced by the reform. The toppanels compare the level of reported earnings, and the bottom panelsdisplay the coefficients from the difference-in-difference regressions onthe two series in the top panels. Appendix Fig. A6 shows similar plotsfor the balanced-panel samples. Collectively, the evidence shows thatthe reported earnings were on parallel trends in the prereform periods.They continued to evolve on the preexisting trend for the control firmsbut declined sharply for the treated firms in 2009. In the post-2009period, the treated earnings began recovering, growing at almost theprereform rate. This, however, means that the tax rate increase caused alasting damage to the tax base: the level of partnership earnings waspermanently lower in the post-reform periods.

Results (2009) — Table 2 reports the results from Eq. (12), re-stricting the sample to the period 2006–09. Starting with the baselinespecification in column (1), columns (2)–(5) successively add morecontrols; columns (5)–(10) replace the year fixed effects with a lineartime trend; Table A2 replicates the exercise on a balanced-panelsample; Table A3 permutes among the combinations of time-trend andbalanced-panel specifications; and finally Table A4 repeats the analysisafter reweighting the two control samples to match the treatmentsample on size and industry dimensions using the DiNardo et al. (1996)method.

Two conclusions emerge from the above analysis. First, firms’earnings choices conditional on participation are extremely elastic tothe tax rate: every percentage point decrease in the net-of-tax rate wasassociated with an almost twice-as-large drop in reported earnings. Thisreflects the small costs at which firms in a low tax-capacity setting areable to manipulate their earnings following a tax change. Second, theelasticity is estimated cleanly, being robust to the identification con-cerns mentioned in Section 4.1. Notably, the results are insensitive to(i) the choice of control group; (ii) holding the composition of thesample fixed; (iii) the choice of time trend (flexible vs. parametric); (iv)comparing firms within an industry, region, and tax-office; (v) allowingfirms in each industry to have a separate growth-trend; (vi) DFL-re-weighting the control samples to match the treatment sample; and (vii)dropping firms affected by income shifting from the control groups(Table 2 vs. Table A5). Furthermore, the placebo coefficient is statis-tically insignificant in almost all the 120 specifications reported inTables 2 and A2 –A6.25 The robustness of the results is a reflection ofthe stability and flatness of the preexisting partnership earnings trend.In fact, the trend was so flat that even the time-series estimates, re-ported in Table A6, are indistinguishable from the corresponding dif-ference-in-differences estimates.

Results (2009–11) — Table 3 reports the results from Eq. (12), se-parately estimating the elasticity in the three post-reform periods. Ap-pendix Fig. A6 display the visual analog of the results, and Table A7presents the corresponding time-series estimates.26 Consistent withFig. 4, the results here show that the intensive-margin response pro-duced by the tax reform was of an immediate and permanent nature:reported earnings underwent a steep decline in 2009 but startedgrowing from this low base at almost the prereform rate after 2009.

24 Specifically, I estimate Eq. (16) on a daily frequency (t = day) for a two year timeperiod from June 6, 2009 to June 5, 2011, defining the last year as the post-reformperiod. I compare δ from this regression against that from placebo regressions, which arerun identically on a two-year window with the last year defined as the post-reform per-iods. The estimation window for the placebo regressions starts from the period June 6,2007 to June 5, 2009 and goes systematically back to July 1, 1995.

25 The insignificance of the placebo coefficients across specifications shows that thereported earnings of treated firms do not change significantly from one year to the nextrelative to the control firms for any nontax reason including mean-reversion. This isconsistent with the graphical evidence showing parallel trends in Fig. 4.

26 Note that one important distinction between the results here and those above is thatthe control samples here are restricted to firms that are not impacted by the 2010 taxchanges (please see discussion in Section 4.1).


50

Reflecting this, the elasticities underlying the post-2009 responses areeither statistically insignificant (Panel B) or negligible relative to the2009 elasticity (Panel A).27 The retroactive application imparts addi-tional significance to the result, implying that the principal channelthrough which firms responded to the rate increase was tax evasion andnot a real change in activity, a point I come back to in Section 5.3 of thepaper.28

5.2.2. Extensive marginGraphical evidence — The three steps of the strategy to estimate the

extensive margin response are displayed in Fig. 5. In the first step, I

estimate the counterfactual number of tax filers in the post-reformperiods. The difference-in-differences setup for this estimation is shownin the first two panels of Fig. 5. The prereform filing trend was in-creasing and reasonably parallel among partnerships and corporations.By contrast, the trend was decreasing for sole proprietorships, implyingthat the extensive margin elasticities using this group of firms as controlwould be underestimated. To account for this, I take two measures.First, I estimate two variants of the baseline model, allowing linear andseparate linear time trends in filing. Second, I supplement the analysiswith within-partnerships comparisons. The basis for this exercise isprovided in Appendix Fig. A7. Panel A of the figure compares thenumbers of partnerships with earnings in the range zit ∈ (0 650 K] andzit ∈ [0 650 K]. The difference between the two numbers for a givenyear represents the partnerships which report zero earnings in the year.The difference was reasonably stable in the prereform periods but grewsharply after the reform, as more firms – compelled by the increase intax rate – shifted to zero earnings. By contrast, such difference for thetwo groups of control firms remained stable throughout the period2006–11 (Panels B and C of the figure). Estimating the counterfactualnumber of tax filers from the two series in Panel A, thus, provides aclean and conservative lower bound on the extensive margin

sgninraEdetropeR:BsgninraEdetropeR:A11

11.5

12

12.5

Average L

og R

eporte

d E

arnin

gs

2006 2007 2008 2009 2010 2011

Partnerships Sole Proprietorships

11

11.5

12

12.5

Average L

og R

eporte

d E

arnin

gs

2006 2007 2008 2009 2010 2011


C: Difference-in-Differences D: Difference-in-Differences

−.6

−.4

−.2

0.2

Diffe

rence−

in−

diffe

rences C

oeffic

ient

2007 2008 2009 2010 2011


−.4

−.2

0.2

Diffe

rence−

in−

diffe

rences C

oeffic

ient

2007 2008 2009 2010 2011


Fig. 4. Intensive margin response. Notes: The figure compares the evolution of reported earnings across partnerships and the two control groups, documenting parallel trends up to thereform and a steep decline in treated earnings thereafter. The top two panels compare the mean of log reported earnings in repeated cross-sections over time, and the bottom paneldisplays the coefficients from the DD regressions on the two series in the top two panels. The sample for the figure contains only firms that report positive earnings in the range zit ∈ (0 650K], thus isolating earnings response conditional on participation created by the reform. The solid vertical line in each panel indicates the time from which the tax changes take effect.

27 An important caveat to these results is that the sample in the three post-reform yearschanges because of the extensive margin response to the reform. It has the potential tointroduce a bias in the estimated coefficients for the three years, although the compar-ability of the estimates from complete samples and balanced-panel samples largely mi-tigates this concern (compare LHS panels of Fig. A6 with the RHS panels, and Table 2 withTable A2).

28 The relative insignificance of the 2010–11 responses compared to the 2009 responsealso suggests that firms use rudimentary, low cost technologies to achieve tax evasion, aswith additional time more sophisticated evasion technologies such as keeping multiplebooks of accounts become feasible.


51

response.29

Using the results from the step one, I next estimate the completecounterfactual distribution of partnership earnings in the post-reformperiods using the assumption Eq. (13). This distribution for the year2009 is shown in Fig. 5A. In the final step, I strip the observed dis-tributions of the intensive-margin responses using the elasticities

estimated in the previous section. This distribution for 2009 is shown inFig. 5B. The difference between the observed and counterfactual dis-tribution that has been stripped of intensive-margin responses isolatesthe extensive margin response to the reform, which I use to estimate theelasticities reported below.

Results — Table 4 summarizes the results. Columns (2)–(4) of thetable report in each row the numbers of firms in the observed andcounterfactual distributions for the given post-reform period t and theincome-weighted elasticity implied by them.30 Columns (5)–(8)

Table 2Intensive margin elasticities (2009).

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

A: Sole proprietorships as controlElasticity 2.233 2.253 2.251 1.999 2.009 2.219 2.241 2.238 1.973 1.981

(0.077) (0.078) (0.077) (0.074) (0.075) (0.079) (0.080) (0.080) (0.077) (0.077)Placebo 0.025 0.028 0.029 0.095 0.100 0.036 0.036 0.038 0.087 0.094

(0.044) (0.044) (0.044) (0.052) (0.052) (0.044) (0.044) (0.044) (0.050) (0.050)Observations 848,466 848,466 811,075 174,475 174,470 848,466 848,466 811,075 174,475 174,450B: Corporations as controlElasticity 1.915 2.112 2.169 1.664 1.893 1.963 2.125 2.240 1.744 1.974

(0.273) (0.280) (0.256) (0.264) (0.255) (0.241) (0.247) (0.210) (0.221) (0.203)Placebo −0.222 −0.212 0.051 0.020 0.212 −0.179 −0.120 0.003 −0.094 0.071

(0.447) (0.485) (0.408) (0.430) (0.426) (0.202) (0.212) (0.177) (0.194) (0.185)Observations 32,722 32,722 32,640 21,338 21,272 32,722 32,722 32,640 21,338 21,272ControlsRegion Fixed Effects No Yes No No Yes No Yes No No YesTax Office Fixed Effects No No Yes No Yes No No Yes No YesIndustry Fixed Effects No No No Yes Yes No No No Yes YesTime Trend Flexible Flexible Flexible Flexible Flexible Linear Linear Linear Linear Linear

Notes: The table presents intensive margin elasticity estimates from Eq. (12) estimated on the period 2006–09. Standard errors are in parenthesis, which are clustered at the firm level.The treatment group comprises partnership firms, and the results in Panel A and B are from using sole proprietorships and corporations as the control group. The estimates in column (1)are from the baseline specification; columns (2)–(5) add additional control variables. I do not observe the industry and tax office for all firms, owing to which the sample for thecorresponding specifications is smaller than that for the others. Columns (6)–(10) replace year fixed effects in Eq. (12) with a linear time trend. Placebo results are from the correspondingspecification estimated on the period 2006–08, with 2008 assumed as the post-reform period. The estimates are income-weighted, so that the elasticity corresponds to the parameter εp inEq. (10).

Table 3Intensive margin elasticities (2009–11).

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

A: Sole proprietorships as controlElasticity (2009) 2.233 2.243 2.240 1.997 2.009 2.162 2.176 2.173 1.833 1.832

(0.077) (0.077) (0.077) (0.075) (0.075) (0.076) (0.077) (0.076) (0.073) (0.074)Elasticity (2010) 0.205 0.251 0.231 0.253 0.275 0.120 0.159 0.142 0.142 0.146

(0.066) (0.067) (0.066) (0.079) (0.080) (0.062) (0.063) (0.062) (0.070) (0.072)Elasticity (2011) 0.025 0.083 0.052 0.099 0.112 0.070 0.128 0.095 0.122 0.139

(0.062) (0.062) (0.063) (0.075) (0.077) (0.061) (0.062) (0.062) (0.074) (0.076)Observations 876,317 876,317 838,773 192,776 192,771 876,317 876,317 838,773 192,776 192,771B: Corporations as controlElasticity (2009) 1.855 2.074 2.149 1.603 1.931 1.855 2.074 2.149 1.603 1.931

(0.316) (0.322) (0.303) (0.310) (0.301) (0.316) (0.322) (0.303) (0.310) (0.301)Elasticity (2010) −0.482 −0.319 −0.242 −0.301 −0.075 −0.482 −0.319 −0.242 −0.301 −0.075

(0.346) (0.350) (0.334) (0.344) (0.337) (0.346) (0.350) (0.334) (0.344) (0.337)Elasticity (2011) 0.352 0.582 0.699 0.591 0.818 0.352 0.582 0.699 0.591 0.818


Notes: The table presents intensive margin elasticity estimates from Eq. (12) estimated on the period 2006–11. Standard errors are in parenthesis, which are clustered at the firm level.The treatment group comprises partnership firms, and the two control groups here are the sole proprietorships and corporations which do not experience the 2010 tax changes (see thediscussion in Section 4.1). I do not observe the industry and tax office for all firms, owing to which the sample for the corresponding specifications is smaller than that for the others.Columns (6)–(10) replace year fixed effects in Eq. (12) with a linear time trend. The estimates are income-weighted, so that the elasticities correspond to the parameter εp in Eq. (10).

29 To see the lower bound interpretation, recall that the extensive margin response tothe reform could potentially occur through three distinct channels: (1) reduced entry; (2)increased exit comprising firms that exit and stop filing returns; (3) increased exit com-prising firm that continue filing returns but report zero earnings. The difference betweenthe two series in Panel A of Fig. A7 captures only the third channel of the response, asboth series are affected by channels (1) and (2) equally.

30 It is important to emphasize that the elasticity for period t here represents the ag-gregate extensive margin response up to the period. In distinction, the period t elasticityreported in the other tables represents the additional response in the period.


52

replicate the exercise but estimate the counterfactual number of taxfilers using the two alternative specifications mentioned above. PanelsA and B use sole proprietorships and corporations as controls in the firststep of the strategy, whereas the estimates in Panel C are from within-partnership comparisons. Table A8, which carries the results from theauxiliary regressions Eq. (14), conducts additional robustness tests.

Three key conclusions emerge from the results. First, consistent withthe visual evidence the extensive margin elasticities are large, reflectingextreme sensitivity of firms’ participation choices to the tax rate.Second, in contrast to the intensive-margin response the extensive-margin response grows over time, as more firms exit and fewer firmsenter the partnership sector. Third, the elasticity estimates, with theexception of ones from the baseline specification in Panel A, are rea-sonably robust across alternative specifications. This conclusion isstrengthened further by the evidence in Appendix Tables A8 to A11,where I compare the propensity to report positive earnings across thethree types of firms over time using different specifications and esti-mation methods. Specifically, the results are robust to (i) holding thecomposition of the sample fixed (Table A9 vs. Table A8); (ii) experi-menting with alternative time trends (Table A9); (iii) conditioning on

more control variables (Table A8); and (iv) using a probit instead of alinear model (Table A10 vs. Table A8).

The 2010–11 elasticity estimates in Table 4 rely on the assumptionthat the extensive margin control outcomes were not impacted by the2010 tax changes. This assumption, as discussed in Section 4.1, is in-nocuous for corporate firms. Table A11, assesses the plausibility of thisassumption for the other control group. Comparing extensive marginoutcomes across sole proprietorships and corporate firms, it shows thatif anything the 2010 tax changes caused a slight reduction in thenumber of sole proprietorships with positive earnings.31 The result,thus, reinforces the conclusion that the extensive margin elasticitiesfrom specifications that use sole proprietorships as counterfactual areunderestimated.

sreliFxaTforebmuN:BsreliFxaTforebmuN:A

10.5

11.5

12.5

13.5

8.5

99.5

10

10.5

Log N

um

ber o

f F

ilers

2006 2007 2008 2009 2010 2011


4.5

5.5

6.5

7.5

8.5

8.5

99.5

10

10.5

Log N

um

ber o

f F

ilers

2006 2007 2008 2009 2010 2011


C: Observed and Counterfactual Distribution D: Observed and Counterfactual Distribution

Δm = −47.1%

0400

800

1200

1600

2000

Num

ber o

f F

ilers

0 100 200 300 400 500 600


Observed (2009) Counterfactual (2009)

Δm = −48.3%

0400

800

1200

1600

2000

Num

ber o

f F

ilers

0 100 200 300 400 500 600


Observed w/o int. response (2009) Counterfactual (2009)

Fig. 5. Extensive margin response. Notes: The figure depicts the strategy to estimate the extensive margin response to the reform. The top two panels compare the number of firms thatreport positive earnings in the range zit ∈ (0 650 K] across the three groups and over time. The solid vertical line in each panel indicates the time from which the tax changes take effect.The bottom two panels illustrates the last two steps of the strategy. Panel C compares the observed and counterfactual partnership earnings distributions in 2009. The counterfactualdistribution is obtained under assumption Eq. (13) by scaling up the 2008 distribution by a factor N N/c

2009 2008, where N c2009, the counterfactual number of tax filers in 2009, is obtained

from the difference-in-differences regression Eq. (12) on the two series in Fig. 5B with a linear time trend (corresponding to the specification in columns (5)–(6) of Table 4). Panel Dcompares the two distributions when the observed distribution has been stripped of intensive responses using the elasticities reported in Table 2. The difference in the number of firms inthe two distributions as a percentage of the number of firms in the counterfactual distribution is denoted by Δm.

31 In fact, this should not be surprising as the most salient 2010 tax change was anincrease in the exemption cutoff, which would necessarily have resulted in a decrease inthe number of tax filers. Though the tax code requires all registered taxpayers, includingthose whose income is below the exemption cutoff, to file a tax return, the tax authoritiesare much less likely to bring action against nonfilers if their tax liability is zero.


53

5.2.3. Income shiftingIncome shifting to sole proprietorships— Fig. 6 plots the distribution of

sole proprietorship earnings reported by former partnership owners,showing that the number of such owners reporting positive sole earn-ings went up considerably after the reform. This visual evidence onincome shifting is formalized in Tables A12 and A13, where I report theresults from regressions similar to Eq. (15). The results shows that therewas significant income shifting towards the sole proprietorship businessform: the propensity to report positive sole earnings by former part-nership owners, columns (3)–(4), underwent a significant, tax-drivenincrease in 2009–11. But the income shifting had relatively little impactin making up for the erosion of the treated base: the tax base loss afteraccounting for the income shifting, columns (5)–(6), is almost as strongas one without it, columns (1)–(2). This happens because the incomeshifting base was so small to start with that even the relatively largeproportional change (the two middle columns) resulted in a smalloverall effect (the difference between the first- and last-two columns).One other reflection of the modest compensatory influence of the in-come shifting is that the revenue-weighted elasticity ηsp, estimated fromEq. (15) and reported in Table 5, is an order of magnitude smaller thanthe corresponding extensive margin elasticity. Table A14 performs ad-ditional specification checks on the results, highlighting their robust-ness to the identification concerns noted in Section 4.1.

Income shifting to corporate firms — Table A15 explores incomeshifting to corporations, reporting results from Eq. (15) with a dummyindicating if agent i reports positive dividend income in period t as theoutcome variable. Table A16 reproduces the results on a balanced panelsample. It is important to reiterate that while income shifting to soleproprietorship could take place from 2009, owing to the fresh regis-tration requirement (see Section 3.2) income shifting to corporationscould only begin from 2010. Reassuringly, the results conform to this,reflecting significant income shifting in 2010 and 2011 but not in 2009.The magnitude of the effect, however, is extremely small: the former

partnership owners’ propensity to report positive dividend income in-creases on average by only 0.3 percentage-points in 2010–11. Thisstatistically significant but otherwise trivial impact is mirrored in therevenue-weighted elasticity ( ηcp) reported in Table 6, which is generallysmaller than one-hundredth of the corresponding extensive marginelasticity.32

Fig. A8, which displays the result from the nonparametric permu-tation test (see Section 4.3), reinforces the conclusion. The estimation ofEq. (16) on daily entry data from June 6, 2009 to June 5, 2011 producesa value of −0.012 for δ with a standard error of 0.007. This suggeststhat the corporate entry did not deviate significantly from the pre-existing trend at the time of the reform. To put the result into per-spective, the permutation test compares it against the distribution of theplacebo coefficient obtained from the equation. The estimated coeffi-cient is right in the middle of the distribution, supporting the aboveresult that the reform had no meaningful impact on the incorporationmargin.33

5.2.4. Spillover effects on the VAT baseTable 7 probes the negative impact of the income tax reform on

government revenue from the value-added tax (see Fig. A9 for thenonparametric evidence). I split the sample into firms subject and not

Table 4Extensive margin elasticities.

Year #Obs. #Counter. Elasticity #Counter. Elasticity #Counter. Elasticity

(1) (2) (3) (4) (5) (6) (7) (8)A: Sole proprietorships as control2009 11,325 16,811 0.924 19,251 1.399 23,243 1.961

(0.026) (0.031) (0.044)2010 8090 16,511 1.717 18,796 2.077 27,401 2.895

(0.038) (0.040) (0.064)2011 6955 14,817 1.743 18,352 2.311 32,301 3.338

(0.039) (0.043) (0.074)B: Corporations as control2009 11,325 20,150 1.545 21,923 1.798 23,243 1.961

(0.032) (0.039) (0.044)2010 8090 17,394 1.867 24,376 2.673 27,401 2.895

(0.039) (0.058) (0.064)2011 6955 17,718 2.226 27,103 3.079 32,301 3.338

(0.041) (0.067) (0.074)C: Within-partnerships comparison2009 11,325 20,741 1.634 – – – –

(0.034) – – – –2010 8090 18,671 2.060 – – – –

(0.040) – – – –2011 6955 17,388 2.179 – – – –

(0.041) – – – –Specification Baseline Linear trend Separate linear trend

Notes: The table presents the extensive margin elasticity estimates from the three-step strategy detailed in Section 4.2. Columns (3), (5), and (7) report in each row the number of tax filersin the counterfactual distribution for the corresponding period t, whereas column (2) reports for the same period the number of tax filers in the observed distribution that has beenstripped of intensive responses. The difference between the two distributions represents extensive margin response to the reform. The number of counterfactual tax filers — the first-stepof the strategy — for estimates in columns (1) to (3) has been estimated using the standard difference-in-differences approach on the filers series depicted in Fig. 5. The estimates in PanelA use sole proprietorships as control, the estimates in Panel B use corporations as control, and the estimates in Panel C use partnerships with earnings in the range zit ∈ [0 650 K] ascontrols (see Fig. A7). For estimates in columns (5)–(8), I add linear and separate linear time trends to the DD specification in the first step of the strategy. The elasticities and standarderrors are from regressions where the difference in the number of tax filers in the two distributions in narrow income ranges is regressed on the change in net-of-participation-tax rateexperienced by taxpayers in the range. The estimates are weighted by taxable income so that the elasticities correspond to the parameter ηp in Eq. (10).

32 This result should not be surprising when seen in light of the costs, both fixed andvariable, that operating a firm as a corporation entails. Corporations for example arerequired to (i) register with the SECP after paying a nontrivial fee; (ii) register with a hostof other federal, provincial, and local departments; (iii) maintain certified audited ac-counts; (iv) comply with labor, social security, and other related regulations; (v) act as taxwithholding agents, deducting tax at source on transactions with other firms. For smallfirms, these costs could easily dwarf the productivity gains or tax savings that mightaccrue from incorporation.

33 I also explore income shifting to the formal wage-earning sector, finding that only afew former partnership owners (less than 20) became wage-earners in periods followingthe tax reform. For all practical purposes, therefore, we can ignore this additional channelthrough which income shifting could occur.


54

subject to VAT and use Eq. (12) and the three-step strategy to estimatethe intensive and extensive margin elasticities for the two groups offirms separately. The results show that although the responses of VAT-registered firms are generally smaller than those of the other firms, theynevertheless are substantial. This creates two mutually reinforcingforces – the component of the tax base not subject to VAT is more elasticand the component subject to VAT though less elastic faces a con-siderably higher tax rate – that will push the efficiency costs of thereform beyond those given by formula (10). In Section 6, I use themethodology described in Appendix A.2 to quantify this additionalwelfare impact, showing that incorporating the fiscal externality raisesthe costs of the reform by nearly 40%.

5.2.5. HeterogeneityTo explore response heterogeneity, I estimate the triple-difference

versions of Eqs. (12) and (14), interacting the double-difference termswith the firm-characteristic variable. I study six firm-characteristics,which proxy for the size, sophistication, and transparency of operationsof a firm (details in Appendix A.1). To avoid making strong functionalform assumptions, the variables are introduced into the equationsnonparametrically as dummies. The results, reported in Tables A17 andA19, show that firms with any of the six characteristics respond con-siderably less aggressively relative to the other firms. Relatedly, TableA18 and Fig. A10 evaluate if the responses vary across income groups orif it is easier for firms in a few industries to become informal.34 Overall,the results are broadly in line with the recent theoretical literature (seefor example Gordon and Li, 2009, Kleven et al., 2016) that emphasizesthe importance of information environment – the extent to which theincome generation process in a firm leaves verifiable information trailsfor the government – as the key determinant of firm compliance.

5.3. Discussion

Real vs. evasion margins — Can governments curtail efficiency costsarising from behavioral responses to taxation? The answer to thequestion depends on the nature of the responses. Real responses are afunction of deep structural parameters of technology and preferences,and therefore are not particularly amenable to policy intervention. Incontrast, governments can always curb evasion and avoidance by im-proving tax design or investing in the enforcement capacity. The esti-mates in Table 3 show that the additional earnings response in 2010–11was negligible relative to the 2009 response. This under unanticipatedretroactive application implies that the predominant channel throughwhich firms responded to the tax increase was tax evasion. This findingis further supported by looking at the evolution of important line itemsreported on the tax return form. Fig. A11 carries out this exercise,showing that the line items more tightly linked to the real side ofbusiness activity, such as profit and loss expenses and inventories (thebottom two panels), do not respond at all in 2009. No change in theseline items – especially in profit and loss expenses which include inputcosts such as wages, rents, utility payments, and thus are more likely tobe third-party reported – compared to the large decrease in taxableincome (Fig. 4), sales (Fig. A11), and cost of sales (Fig. A11) supportsthe tax evasion interpretation of the observed 2009 response.

Unlike the intensive response, I am less certain about the exactnature of the extensive margin response. Table 4 shows that the tax rateincrease led to fewer and fewer formal firms over time. While some ofthe missing firms would have migrated to the informal sector, theothers would have shut down completely. Unfortunately, the two ex-tensive margin choices that I observe in the data – firms reporting zeroearnings or disappearing completely after the reform – are potentiallyconsistent with both explanations. But, as noted earlier, considering thelack of public support available to the exited taxpayers through socialinsurance, it would be natural to consider that the extensive response inlarge part reflects exit into informality.

External validity — Though the empirical focus of this paper arepartnerships, for at least two reasons the results are broadly re-presentative of firm behavior to taxation in a low enforcement-capacitysetting. First, in terms of their reaction to taxes partnerships are quitesimilar to the other types of firms. Table A20 establishes this formallyby comparing the behavior of sole proprietorships and partnerships tothe common baseline tax system in 2006–08. Furthermore, in terms of

A: Sole Proprietorship Income of Partnership Owners – Prereform

Δm2007 = −3.8%

Δm2008 = −1.5%

02

00

40

06

00

80

0

Nu

mb

er o

f F

ile

rs

0 100 200 300 400 500 600

Sole Proprietorship Income in PKR 000s

2006 2007 2008

B: Sole Proprietorship Income of Partnership Owners – Post-reform

Δm2009 = 17.9%

Δm2010 = 17.2%

Δm2011 = −3.9%

02

00

40

06

00

80

0

Nu

mb

er o

f F

ile

rs

0 100 200 300 400 500 600

Sole Proprietorship Income in PKR 000s

2008 2009 2010 2011

Fig. 6. Income shifting to sole proprietorships. Notes: The figure plots sole proprietorshipearnings reported by former partnership owners from 2006 to 2011, exploring therebyincome shifting towards the business form. Partnership owner here is defined as an in-dividual who reports earnings from a partnership in any of the three prereform periods2006–08. Each dot on the curves represents the upper bound of a PKR 10,000 bin anddenotes the number of partnership owners who report sole earnings within that bin.While plotting the 2010–11 distributions, I partial out the intensive responses to the 2010tax changes using Eq. (13). Yearly changes in the number of tax filers are shown by Δmt,which for year t signifies the change in the number of filers from year t−1 to t as apercentage of the number of filers in year t−1.

34 The analysis shows that the extensive response to the reform was generally homo-geneous across industries with the exception of a few outliers. These outlier industrieswere Retail Sales; Manufacture of plastic, concrete, and plaster products; Repair ofMachinery, Equipment, and Vehicles (stronger than usual response) and Real EstateServices, Accommodation Services, and Construction of Buildings (no negative responseat all).


55

Table 5Income shifting to sole proprietorships.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

A: Sole proprietorships as controlElasticity (2009) −0.211 −0.212 −0.212 −0.146 −0.154 −0.188 −0.201 −0.198 −0.142 −0.156

(0.006) (0.006) (0.006) (0.011) (0.011) (0.005) (0.005) (0.005) (0.010) (0.010)Elasticity (2010) −0.255 −0.261 −0.257 −0.185 −0.200 −0.280 −0.287 −0.282 −0.216 −0.233

(0.004) (0.004) (0.004) (0.008) (0.008) (0.005) (0.004) (0.004) (0.009) (0.008)Elasticity (2011) −0.323 −0.325 −0.321 −0.235 −0.250 −0.306 −0.304 −0.303 −0.233 −0.244

(0.004) (0.004) (0.004) (0.008) (0.008) (0.004) (0.004) (0.004) (0.008) (0.008)Observations 2,991,995 2,991,995 2,818,285 867,567 867,551 2,991,995 2,991,995 2,818,285 867,567 867,551B: Corporations as controlElasticity (2009) −0.143 −0.147 −0.149 −0.142 −0.142 −0.137 −0.139 −0.143 −0.138 −0.138

(0.008) (0.008) (0.008) (0.011) (0.011) (0.006) (0.006) (0.006) (0.010) (0.010)Elasticity (2010) −0.267 −0.282 −0.270 −0.262 −0.277 −0.230 −0.227 −0.233 −0.229 −0.229

(0.008) (0.008) (0.008) (0.010) (0.011) (0.006) (0.006) (0.006) (0.009) (0.009)Elasticity (2011) −0.237 −0.229 −0.238 −0.238 −0.233 −0.249 −0.243 −0.251 −0.253 −0.251


Notes: The table presents the income shifting elasticity estimates from Eq. (15), estimated on the period 2006–11. Standard errors are in parenthesis, which are clustered at the firm level.The treatment group comprises partnership owners, defined as individuals who report earnings from a partnership in any of the three prereform periods 2006–08. The outcome variable isan indicator 1(zj,it>0) denoting that i reports positive sole proprietorship earning (taxable income for corporations) in period t. I do not observe the industry and tax office for all firms,owing to which the sample for the corresponding specifications is smaller than that for the others. The estimates are revenue-weighted so that the elasticities correspond to the parameter

ηsp in Eq. (10).

Table 6Income shifting to corporations.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

A: Sole proprietorships as controlElasticity (2010) −0.013 −0.011 −0.011 −0.023 −0.016 −0.012 −0.010 −0.010 −0.019 −0.012

(0.003) (0.003) (0.003) (0.007) (0.007) (0.004) (0.003) (0.003) (0.007) (0.007)Elasticity (2011) −0.015 −0.013 −0.013 −0.031 −0.024 −0.015 −0.015 −0.015 −0.029 −0.023

(0.004) (0.004) (0.004) (0.007) (0.007) (0.004) (0.004) (0.004) (0.007) (0.007)Observations 2,991,995 2,991,995 2,818,285 867,567 867,551 2,991,995 2,991,995 2,818,285 867,567 867,551B: Corporations as controlElasticity (2011) −0.022 −0.027 −0.024 −0.021 −0.025 −0.018 −0.022 −0.020 −0.020 −0.022

(0.004) (0.005) (0.004) (0.008) (0.008) (0.004) (0.004) (0.004) (0.008) (0.008)Elasticity (2010) −0.026 −0.035 −0.029 −0.031 −0.035 −0.024 −0.035 −0.029 −0.033 −0.039


Notes: The table presents the income shifting elasticity estimates from Eq. (15), estimated on the period 2006–11. Standard errors are in parenthesis, which are clustered at the firm level.The treatment group comprises partnership owners, defined as individuals who report earnings from a partnership in any of the three prereform periods 2006–08. The outcome variable isan indicator 1(zj,it>0) denoting that i reports positive dividend income in period t. I do not observe the industry and tax office for all firms, owing to which the sample for thecorresponding specifications is smaller than that for the others. Columns (6)–(10) replace year fixed effects in Eq. (12) with a linear time trend. The estimates are revenue-weighted so thatthe elasticities correspond to the parameter ηcp in Eq. (10).


56

their compliance attributes partnerships lie in between the other twotypes of firms (Table 1), making their behavior typical of an averagefirm in the country. Second, partnerships in Pakistan are not limited tohuman-capital intensive industries such as accounting and law but arerepresented fairly in all industries (Fig. A1).35

Relatedly, it is commonly known in literature that earnings re-sponses to a tax reform depend on its design, in particular the incomeshifting opportunities it affords (see for example Slemrod and Kopczuk,2002). The reform exploited here was targeted to a narrow section ofthe tax base, thus creating significant opportunities for income shifting.This, however, does not diminish the external validity of the results, as Iam able to identify income shifting cleanly. Once these fiscal ex-ternalities are netted out, the leftover responses characterize whatwould occur if a similar but broad-based tax increase is implemented.

6. Welfare analysis

In this section, I use formula (10) to compute the welfare costs of thereform. To express these costs in more intuitive terms, I compute thefollowing statistic

− =

+ + +−

dW dτdW dτ

τ ε η η ητ

// |

( )1

,p

p z

p p p sp cp

pp (17)

where dWdτ zp

is the change in welfare absent any behavioral response,

commonly known in the literature as the mechanical effect of a taxchange. The statistic expresses the behavioral revenue loss caused bythe reform as a share of its mechanical effect. If this share is larger thanone, the government would lose more revenue than the maximum it

could have gained from the tax increase, implying that the new tax ratewas not optimal. Using the most conservative estimates of the fourelasticities estimated in the last section, I obtain a value of 1.25 for thestatistic in 2011. This illustrates that the new, flat tax rate of 25% onpartnership earnings was on the wrong side of the Laffer curve andwould not have been optimal under any social preferences.36 Thiscomputation ignores the negative externality operating through thereduction in VAT revenue. Using the methodology developed in Ap-pendix A.2 and the elasticities reported in Table 7, I find that the valueof the statistic rises by more than 40% once the additional impact istaken into account.

An alternative, and perhaps more transparent, way to show that thenew rate of 25% was above the Laffer bound is to compare the pre-dicted post-reform revenue under the baseline tax system to the actualrevenue realized after the reform. The counterfactual partnershipearnings distributions estimated in Section 5.2.2 (see Fig. 5) allow meto predict the revenue the government would have obtained in 2009–11had it not changed the tax system. I find that the predicted revenue isstrictly larger than the actual revenue realized in 2011, showing thatthe tax rate increase led to a decrease rather than an increase in rev-enue. I provide the details of the methodology used for this exercise inAppendix A.3.

7. Conclusions

Firm behavior to taxation in settings characterized by low en-forcement capacity and large informality has been understudied, pri-marily because the relevant data were not available till recently. Thispaper uses the population of income tax returns filed between 2006 and2011 in Pakistan to show how firms engage in tax evasion, migrate intoinformality, and switch business organization to counter an increase intheir tax burden. Elasticities underlying these responses are an order ofmagnitude larger than ones estimated in rich countries, highlighting thesmall costs at which firms in developing countries are able to manip-ulate their earnings following a tax increase.

The identifying variation in the paper comes from a policy reformintroduced in 2009. The Pakistani context offers three key advantages.First, the reform creates tax rate variation across very similar firms,thereby producing almost ideal comparison groups to disentanglemacroeconomic shocks from the tax-driven responses. Second, the re-form was given retroactive effect that helps decomposing the observedresponses into real and evasion margins. Third, the context, in parti-cular the richness of the data, permits a clean identification of key fiscalexternalities – income shifting and negative impact on the VAT base –arising from the change in marginal income tax rate. Taking the spil-lovers into account leads to more robust measurement of welfare lossarising from the tax change.

The results underscore that informality and tax evasion remain thefirst-order challenges stifling the development of fiscal capacity inemerging economies. Following the tax increase, the number of formal,treated firms fell sharply (elasticity ≈ 3), and surviving firms’ earningsregistered a steep decline (elasticity ≈ 2). Though, some of the losseswere offset by income shifting within the formal sector, the size of thecountervailing effect was extremely modest. Accordingly, the over-riding conclusion that one draws from the results is that unless the costsof noncompliance are increased markedly, through investment in en-forcement capacity or a change in tax design, raising taxes in emergingeconomies would continue to entail crippling economic costs.

Table 7Spillover effects on the VAT base.

Firms subject to VAT Firms not subject to VAT

(1) (2) (3) (4) (5) (6)A: Sole proprietorships as control1. Intensive Margin:Elasticity 1.424 1.364 1.254 2.681 2.676 2.690

(0.130) (0.133) (0.136) (0.096) (0.099) (0.098)Observations 51,556 51,556 7074 796,910 796,910 22,4172. Extensive

Margin:Elasticity 0.677 1.499 2.569 2.121 2.626 3.556

(0.033) (0.034) (0.071) (0.045) (0.048) (0.073)Observations 51,556 51,556 7074 796,910 796,910 22,417B: Corporations as control1. Intensive Margin:Elasticity 1.241 1.102 1.249 2.238 2.400 2.690

(0.496) (0.425) (0.136) (0.288) (0.240) (0.098)Observations 8203 8203 7073 24,519 24,519 22,4172. Extensive

Margin:Elasticity 0.363 1.974 2.490 2.787 3.430 3.551

(0.041) (0.051) (0.068) (0.053) (0.070) (0.073)Observations 8203 8203 7074 24,519 24,519 22,417Specification DD DD TS DD DD TSTime Trend Flexible Linear Flexible Flexible Linear Flexible

Notes: The table replicates the analysis in Tables 2 and 4, stratifying the sample by VATregistration of firms. The intensive margin estimates in columns (1)–(2) and (5)–(6) arebased on Eq. (12), and correspond to the estimates in columns (1) and (6) of Table 2. Theextensive margin estimates in columns (1)–(2) and (5)–(6) are from the three-stepstrategy, and correspond to the estimates in columns (4) and (6) of Table 4. The estimatesin column (3) and (6) are the time series counterparts, and correspond to the relatedestimates in columns (1) of Table A6 and (8) of Table 4. The standard errors are inparenthesis, which are clustered at the firm level. All estimates are income-weighted, sothat the elasticities correspond to the parameters εp and ηp in Eq. (10).

35 A more or less similar trend is observed in the US. See Cooper et al. (2015) for thedistribution of tax-paying firms in the US stratified by their organizational form.

36 This conclusion remains unaltered even if we take into account the marginal effi-ciency gains accruing from the replacement of the notch-based, prereform tax schedulewith the flat, post-reform tax schedule.


57

Appendix A

A.1. Details of firms characteristics

(i) Firm size. The dummy variable takes the value 1 for firms with sales above the 75th percentile of the size distribution,37 where firm size isdefined as the average annual sales in the three prereform periods.

(ii) Electronic return filer. All partnerships were required to file electronic returns in years 2008–11. Some of the firms did not comply with themandatory provision, while a few were filing electronically even before the mandate came into effect. I categorize a firm electronic filer if any ofthe four returns for tax years 2006–09 was filed electronically (about 80% of the firms).

(iii) Registered for VAT. The variable indicates if the firm was registered with the FBR to remit VAT on its sales.(iv) Firm age. The dummy variable takes the value 1 if age of the firm –measured in the number of years since registering with the FBR – was more

than the 75th percentile (6 years).(v) Tax withholding. Pakistani tax code stipulates a very elaborate tax withholding scheme. In addition to wages, tax is withheld on a number of

other transactions including the payment for goods and services, utility bills, cash withdrawal from banks, and imports from other countries.The withheld tax can be adjusted against the tax liability at the time of filing of returns. The firms which withhold tax are required to file astatement with the FBR indicating the transactions and the tax withheld thereon. The scheme has some elements of third party reporting, thoughit does not provide information on the total tax base as is the case with tax withholding on wages. The dummy variable takes the value 1 if thewithheld tax of a firm weighted by its taxable income was more than the 75th percentile in the prereform periods.

(vi) Withholding agent. The variable is an indicator if the firm was a withholding agent, required by the tax code to withhold tax on transactionsmade with its buyers or sellers (about 22% of the firms).

A.2. Spillover effect on the VAT base

Formula (10) ignores one important feature of the tax environment that a subset of firms also remit VAT on their sales. The income-tax-drivenchanges in firm behavior will impact government revenue from the VAT base as well, increasing the costs of the reform above those given by theformula. To incorporate this fiscal externality into the welfare calculations, note that the government revenue in this general setting is given by

∫ ∫∑=⎡

⎣

⎢⎢

+ +⎤

⎦

⎥⎥∈ ∈ ′

τ θ τ θ θ τ θ θT τ τ z dF τ z dF( , ) ( ) ( , ) ( ) ( , ) ( ) ,θ τ θ τj M

j jv

jM

j j( ) ( )j

vjv (A1)

where τM ( )jv and ′ τM ( )j

v are the two mutually exclusive subsets of Mj(τ) consisting of firms of type j which are subject and not subject to VAT, zj arereported earnings, and τj

v is the effective VAT rate on these earnings. The above expression shows that as long as the tax change dτp does not cause amovement of firms from M (.)j

v to ′M (.)jv its welfare impact can be computed simply as the weighted sum of the impact on the two constituent bases.38

Note that this is a very general formulation that allows the two types of firms to have different elasticities, which is important because VAT-registeredfirms are linked to their supplies and buyers through the invoice-credit mechanism and thus might have lower ability to manipulate their earningsafter a tax change. In the empirical application, I therefore compute the welfare costs of the reform in two iterations. I first use formula (10), ignoringthereby the negative VAT externality. I then take the externality into account, and use the elasticities from Table 7 to compute the aggregate welfareloss as a weighted average of the loss in the bases subject and not subject to VAT.

A.3. Was the post-reform tax rate above the Laffer bound ?

In Section 6, I use formula (17) to show that the new tax rate of 25% was on the wrong side of the Laffer curve. In this section, I derive the resultusing a more intuitive and transparent method.

In Section 5.2.2, I construct the counterfactual distributions of partnership earnings in the post-reform periods. Using the distributions, I predictthe counterfactual revenue that would have been realized in a given post-reform year had the tax system remained unchanged. Given that thecounterfactual distributions are estimated in bins of PKR 10 K, I use the approximation that the average income within a bin is the mid-point of thebin. For example, I treat reported income of all firms in the (0 10,000] bin as PKR 5000. Multiplying the average income with the number of firmsand the baseline tax rate gives me the counterfactual revenue in a particular bin. I then aggregate the revenue from the entire binned distribution forthe year. This counterfactual revenue for years 2009–11 is shown in the second column of the following table. I compare it to the revenue actuallyrealized in these years in the third column. Columns (4)–(5) make similar comparisons but also take into account VAT remitted by the firms.39 Thebehavioral revenue loss caused by increasing the tax rate to 25% was so large that by the third year (second year if we take VAT into account) afterthe reform the government was collecting less revenue than it would have under the baseline tax rates. This clearly shows that the new tax rate wasset above the Laffer bound.40

37 The cutoff choice reflects the strongly skewed firm size distribution. The 75th percentile corresponds to an annual turnover of Rs. 6.6 million (US $ 62,857). Compared with this themedian firm has a turnover of Rs. 1.9 million (US$ 17,749) only.

38 In Pakistan, firms subject to VAT are separately registered and are not allowed to drop out of the VAT net unless their sales, verified through a process of detailed audit, fall below theexemption cutoff (Rs. 5 million). For this reason, de-registration from VAT is generally a very costly process and the assumption that firms do not move across the two sets is a gooddescription of the empirical setting.

39 To compute the VAT revenue, I make the very conservative assumption that the value-added of a firm – sales minus the cost of raw materials – is twice its taxable income.40 Note that it is quite a conservative assessment, as in addition to the income tax and value-added tax the government would also lose other small federal, provincial, and local taxes

that are recovered from registered firms.


58

Table A1The counterfactual and realized revenue.

Without VAT With VAT

Year Counterfactual Realized Counterfactual Realized(1) (2) (3) (4) (5)2009 308 581 723 7992010 363 465 852 6482011 428 409 1005 573

All figures are in PKR millions.

A) Industry (Partnerships Vs. Sole Props.) B) Industry (Partnerships Vs. Corporations)

010

20

30

40

50

Percen

tag

e o

f F

irm

s in

th

e In

du

str

y

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25


010

20

30

40

Perce

nta

ge

of F

irm

s in

th

e In

du

str

y

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25


C) Geographic (Partnerships Vs. Sole Props.) D) Geographic (Partnerships Vs. Corporations)

010

20

30

40

Percenta

ge o

f F

irm

s in t

he T

ax D

istr

ict

Lahore

Kara

chi

Multan

Faisalabad

Rawalpindi

Sialkot

Gujranwala

Peshawar

Hydera

bad

Islam

abad

Sukkur

Abbottabad


020

40

60

Percenta

ge o

f F

irm

s in t

he T

ax D

istr

ict

Lahore

Kara

chi

Multan

Faisalabad

Islam

abad

Sialkot

Rawalpindi

Peshawar

Gujranwala

Hydera

bad

Sukkur

Abbottabad


E) Firm Size (Partnerships Vs. Sole Props.) F) Firm Size (Partnerships Vs. Corporations)

02

46

Percenta

ge o

f F

ilers

0 2.5 5 7.5 10

Annual Sales in PKR Millions


01

23

Percenta

ge o

f F

ilers

0 2.5 5 7.5 10

Annual Sales in PKR Millions


Fig. A1. Industry, geographic, and size distribution of firms. Notes: The figure shows the industry, geographic, and size distribution of firms in the baseline year 2008. The top two panelscompare the distribution of firms across the top 25 industries in Pakistan. The detailed description of these 2-digit industries is given in Table A21. The number on the x-axis correspondsto the industry label (column 1) in the table. The middle panels compare the distribution of firms across major cities in Pakistan. The lengths of blue and red bars in panels A–D show theproportion of each type of firm in the particular industry or city. For example Industry 1 contains around 30% of all sole proprietorships and around 20% of all partnerships in Pakistan.The bottom two panels compare the distribution of firm-size across the three types of firms. Each dot in the plots represents the upper bound of a PKR 100 K bin and denotes thepercentage of firms with annual sales within that bin.


59

A) Entry of New Firms

010000

20000

30000

40000

Num

ber o

f N

ew

Entr

ants

−−

Sole

Proprie

torship

s

01000

2000

3000

4000

5000

6000

Num

ber o

f N

ew

Entr

ants

−−

Partn

ership

s

2006 2007 2008 2009 2010 2011 2012


B) Difference-in-Differences

−6

−4

−2

02

Diffe

rence−

in−

Diffe

rences C

oeffic

ient

2006 2007 2008 2009 2010 2011 2012


Fig. A2. Was the reform anticipated? Notes: The figure replicates the analysis in Fig. 2, using sole proprietorships as controls. For sole proprietorships, I observe the date of registrationonly if the firm files a return, as they are not required to register separately from their owners. The analysis in this figure is, accordingly, limited to the subset of firms which file tax returnin the sample period. For this reason, the two entry series decline mechanically over time and are more noisy. The results nonetheless are consistent with those in Fig. 2 which usecorporate firms as control.


60

B) 2007A) 20060

200

400

600

800

10

00

Nu

mb

er o

f F

ile

rs

−5 0 5 10 15 20 25

Marginal Income Shifting Cost (%)

0200

400

600

800

10

00

Nu

mb

er o

f F

ile

rs

−5 0 5 10 15 20 25


D) 2009C) 2008

0200

400

600

800

10

00

Nu

mb

er o

f F

ile

rs

−5 0 5 10 15 20 25


0200

400

600

800

10

00

Nu

mb

er o

f F

ile

rs

−5 0 5 10 15 20 25


F) 2011E) 2010

02

00

40

06

00

80

01000

Nu

mb

er o

f F

ile

rs

−5 0 5 10 15 20 25


02

00

40

06

00

80

01000

Nu

mb

er o

f F

ile

rs

−5 0 5 10 15 20 25


Fig. A3. Distribution of income shifting costs. Notes: The figure depicts the distribution of income shifting costs. The variable on the x-axis captures the minimum profit gain a firmexperiences from operating as a partnership rather than a sole proprietorship. If the firm changes its business organization after the tax increase, it will lose at least this much of profits.The variable, thus, represents a lower bound on the income shifting costs. The details on how these costs are calculated are in Section 3.5. The distributions are shown in bins of size 0.83,where each bin includes the upper bound of the interval. The solid vertical line demarcates the boundary below which such costs are negative. It is important to emphasize that the threepost-reform panels are not directly comparable to the three prereform panels, as some partnerships drop out of the sample in 2009–11.


61

0.2

.4.6

.81

Cu

mu

lative

Dis

trib

utio

n F

un

ctio

n

0 100 200 300 400 500 600


2006 2007 2008

Fig. A4. Empirical earnings distribution function. Notes: The figure plots the Cumulative Distribution Function (CDF) of partnership earnings in Pakistan for the three prereform periods2006–08. The three curves of the figures plot for each year t the probability Pr [zp,it ≤ x] as the cutoff x is varied from 0 to 650 K in intervals of 10 K. That these curves are almost on top ofeach other provides a direct test in support of Eq. (13).

A) Sole Proprietorships – Prereform B) Sole Proprietorships – Post-reform

Δm07 = −6.2%

Δm08 = −4.4%

020

40

60

80

Nu

mb

er o

f F

ile

rs 0

00

s

0 100 200 300 400 500 600


2006 2007 2008

Δm09 = −6.1%

Δm10 = −2.3%

Δm11 = −13.7%0

20

40

60

80

Nu

mb

er o

f F

ile

rs 0

00

s

0 100 200 300 400 500 600


2008 2009 2010 2011

C) Sole Proprietorships – Prereform D) Sole Proprietorships – Post-reform

Δm07 = −6.2%

Δm08 = −4.4%

02

04

060

80

Nu

mb

er o

f F

ile

rs 0

00

s

0 100 200 300 400 500 600


2006 2007 2008

Δm09 = −5.9%

Δm10 = −2.4%

Δm11 = −13.6%

020

40

60

80

Nu

mb

er o

f F

ile

rs 0

00

s

0 100 200 300 400 500 600


2008 2009 2010 2011

Fig. A5. Taxable income distribution. Notes: While plotting sole proprietorship earnings distribution in Fig. 3C–D, I (i) drop sole proprietors that report any income from a partnership in2006–11 and (ii) strip the 2010–11 distributions of intensive responses to the 2010 tax changes using Eq. (13). This figure depicts the sole proprietorship earnings distribution withoutmaking these changes. The top panels replicate panels C and D of Fig. 3 without making the change (i), and the bottom panels replicate panels C and D of Fig. 3 without making thechange (ii). Each dot on the curves represents the upper bound of a PKR 10,000 bin and denotes the number of firms which report earnings within that bin. The notches in the 2006–08schedule are shown by the vertical dotted lines. In the right-hand side panels, the 2008 distribution is shown again for comparison purposes. Yearly changes in the number of tax filers areshown by Δmt, which for year t signifies the change in the number of filers from year t−1 to t as a percentage of the number of filers in year t−1.


62

B) Baseline (BP)A) Baseline−

.6−

.4−

.20

.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


C) With Region Fixed Effects D) With Region Fixed Effects (BP)

−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


E) With Industry Fixed Effects F) With Industry Fixed Effects (BP)

−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


Fig. A6. Robustness of intensive margin estimates. Notes: The figure portrays the robustness of the intensive margin estimates. Each panel of the figure illustrates the results from thefollowing regressions similar to Eq. (12)

= + + + +× β X δz α λ uPartnership YearΔ log ,it it i t it

where Partnership × Year is a vector of three interaction dummies one for each year 2007 to 2011. Each panel of the figure plots the coefficients and the corresponding 95% confidenceintervals on the five interaction dummies. Note that I have to drop the partnership dummy in order to show the double-interaction coefficients for the five periods, so the results thoughvery similar are not directly comparable to those in Table 3. The LHS panels include all firms, whereas the RHS panels show the results for the same specification estimated on thebalanced panel sample. The sample for the estimation include both partnerships and sole proprietorships, and all estimates are weighted by taxable income. The baseline specifications inpanels A–B do not include any control variables. The specification in panels C–H include region, industry, and both region and industry fixed effects. The specifications in panels I–Lreplace the year fixed effects with the linear and industry-specific time trends. The industry-specific time trend specification includes industry, year, and industry × year fixed effects.


63

G) With Region and Industry Fixed Effects H) With Region and Industry Fixed Effects (BP)−

.6−

.4−

.20

.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


J) Linear Trend (BP)I) Linear Trend

−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


L) Industry-specificTrend (BP)K) Industry-specific Trend

−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011


−.6

−.4

−.2

0.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d E

arn

ing

s

2007 2008 2009 2010 2011



64

A) Partnerships (Positive Filers Vs. All)

8.5

99.5

10

10.5

11

Log N

um

ber o

f F

ilers

2006 2007 2008 2009 2010 2011

zit ∈ (0 650K] zit ∈ [0 650K]

B) Sole Props. (Positive Filers Vs. All)

11.5

12

12.5

13

13.5

14

Log N

um

ber o

f F

ilers

2006 2007 2008 2009 2010 2011

zit ∈ (0 650K] zit ∈ [0 650K]

C) Corporations (Positive Filers Vs. All)

56

78

910

Log N

um

ber o

f F

ilers

2006 2007 2008 2009 2010 2011

zit ∈ (0 650K] zit ∈ [0 650K]

Fig. A7. Extensive margin response. Notes: The figure depicts the extensive margin response to the reform. The three panels compare the number of firms that report earnings in therange zit ∈ (0 650 K] to the number of firms that report earnings in the range zit ∈ [0 650 K] across the three types of firms. The difference between the two series in each panel capturesfirms that report zero earnings in a given year. For treated firms, the difference was remarkably stable in the prereform years but grew sharply afterwards. For control firms, the differenceremained stable through out the period 2006–11. The solid vertical line in each panel indicates the time from which the tax changes take effect.


65

0.2

.4.6

.81

Em

piric

al C

DF

−.1 0 .1 .2 .3

Estimated Placebo Coefficient

Fig. A8. Distribution of placebo coefficients. Notes: The figure displays the results of the nonparametric permutation test detailed in Section 4.3. The solid vertical line indicates thecoefficient δ from Eq. (16) estimated on daily entry data (t=day) from June 6, 2009 to June 5, 2011, with the last year defined as the post-reform period. The solid, red curve plots thedistribution of the placebo coefficient from the equation estimated on the prereform periods only. To obtain these placebo coefficients, I begin with the period June 6, 2007 to June 5,2009, defining the last year as the post-reform period, and move systematically backwards up to July 1, 1995 in steps of one week.

A) Intensive Margin

Δlog(z)VAT = −.28

Δlog(z)Others = −.71

−.6

−.4

−.2

0.2

Average L

og C

hange in R

eporte

d E

arnin

gs

2007 2008 2009

VAT−Registered Not VAT−Registered

B) Extensive Margin

ΔmVAT = −18.42%

ΔmOthers = −47.44%

05000

10000

15000

20000

Num

ber o

f F

ilers −

Not V

AT

−R

egis

tered

02000

4000

6000

8000

Num

ber o

f F

ilers −

VA

T−

Regis

tered

2006 2007 2008 2009 2010 2011

VAT−Registered Not VAT−Registered

Fig. A9. Spillover effects on value-added tax base. Notes: The figure replicates the analysis in Figs. 4 and 5, stratifying the sample by VAT registration of a firm. For space considerations,curves related to the treatment group (partnerships) only are shown. Panel A displays earnings growth path of the two group of firms in 2006–09. To help the comparison, average logchange in reported earnings between 2008 and 2009 has been shown in the figure separately for the two types of firms. Panel B illustrates the number of partnerships that report positiveearnings in a given year t. Again, to help comparison, change in the number of firms from 2008 to 2009 as a percentage of the number of firms in 2008 are shown in the figure.


66

02

00

04

00

06

00

08

00

0

Nu

mb

er o

f P

artn

ersh

ips in

th

e I

nd

ustr

y

−.5 −.25 0 .25

Coefficient on the Partnership x Year x Industry Interaction

2009 2010 2011

Fig. A10. Extensive margin response by industry. Notes: The figure explores heterogeneity in the extensive response across industries. I estimate the triple-difference version of Eq. (14),including industry and Partnership × Year × Industry dummies into the specification. The figure plots the distribution of the coefficients on the triple-interaction-dummies for the threepost-reform years, showing that the response was fairly homogeneous with the exception of a few outliers. These outlier industries are Retail Sales; Manufacture of plastic, concrete, andplaster products; Repair of Machinery, Equipment, and Vehicles (coefficient <−0.35) and Real Estate Services, Accommodation Services, and Construction of Buildings (coeffi-cient> 0).

B) Cost of SalesA) Sales

−.3

−.2

−.1

0.1

.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d S

ale

s

2007 2008 2009


−.3

−.2

−.1

0.1

.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Re

po

rte

d C

osts

2007 2008 2009

Partnership Sole Proprietorships

C) Profit and Loss Expenses D) Inventories

−.3

−.2

−.1

0.1

.2

Ave

ra

ge

Lo

g C

ha

ng

e in

Pro

fit

& L

oss E

xp

en

se

s

2007 2008 2009


−.3

−.2

−.1

0.1

.2

Ave

ra

ge

Lo

g C

ha

ng

e in

In

ve

nto

ry

2007 2008 2009


Fig. A11. Real vs. evasion response. Notes: The figure plots the evolution of individual line items reported on the tax return form between 2006 and 2009. Treatment group here arepartnership firms, and the control group are sole proprietorships. The line items considered are the standard items in a profit and loss account. Specifically, Sales represents the moneyreceived in lieu of goods sold and/or services provided; Cost of Sales represents the direct cost of making those sales; Profit and Loss Expenses include wages, rents, utility payments, legaland administrative fee; and Inventories are the opening stock on day one of period t. Each marker on the curves denotes average, within-firm log change in the line item between the yearst−1 and t for the corresponding group of firms. The solid vertical line in each panel indicates the time from which the tax changes take effect.


67

Table A2Intensive margin elasticities (2009) — balanced panel.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)


(0.071) (0.071) (0.071) (0.083) (0.084) (0.071) (0.075) (0.075) (0.089) (0.089)Placebo 0.031 0.033 0.035 0.125 0.130 0.031 0.052 0.053 0.121 0.126


(0.352) (0.360) (0.353) (0.351) (0.358) (0.231) (0.242) (0.228) (0.238) (0.255)Placebo −0.783 −0.847 −0.793 −0.747 −0.825 −0.089 −0.132 −0.095 −0.079 −0.139

(0.369) (0.382) (0.363) (0.377) (0.428) (0.184) (0.194) (0.182) (0.191) (0.210)Observations 9603 9603 9594 6810 6801 9603 9603 9594 6810 6801ControlsRegion Fixed Effects No Yes No No Yes No Yes No No YesTax Office Fixed Effects No No Yes No Yes No No Yes No YesIndustry Fixed Effects No No No Yes Yes No No No Yes YesTime Trend Flexible Flexible Flexible Flexible Flexible Linear Linear Linear Linear Linear

Notes: The table replicates the results in Table 2 on a balanced panel sample containing firms that report positive earnings in all periods included in the estimation 2006–09. Standarderrors are in parenthesis, which are clustered at the firm level. The treatment group comprises partnership firms, and the results in Panels A and B are from using sole proprietorships andcorporations as the control group. The estimates in column (1) are from the baseline specification; columns (2)–(5) add additional control variables. I do not observe the industry and taxoffice for all firms, owing to which the sample for the corresponding specifications is smaller than that for the others. Columns (6)–(10) replace year fixed effects in Eq. (12) with a lineartime trend. Placebo results are from the corresponding specification estimated on the period 2006–08, with 2008 assumed as the post-reform period. The estimates are weighted bytaxable income so that the elasticity corresponds to the parameter εp in Eq. (10).

Table A3Intensive margin elasticities (2009) — robustness I.

(1) (2) (3) (4) (5) (6) (7) (8)

A: Sole proprietorships as controlElasticity 2.233 2.219 2.219 2.193 2.189 2.129 2.153 2.236

(0.077) (0.071) (0.079) (0.075) (0.112) (0.117) (0.079) (0.092)Placebo 0.025 0.031 0.036 0.051 0.036 0.051 0.125 0.131

(0.044) (0.049) (0.044) (0.049) (0.044) (0.049) (0.051) (0.068)Observations 848,466 538,560 848,466 538,560 848,466 538,560 174,475 104,421B: Corporations as controlElasticity 1.915 2.347 1.963 2.019 2.187 2.129 1.520 2.103

(0.273) (0.352) (0.241) (0.231) (0.112) (0.117) (0.306) (0.410)Placebo −0.222 −0.783 −0.179 −0.089 −0.179 −0.089 0.003 −0.938

(0.447) (0.369) (0.202) (0.184) (0.202) (0.184) (0.464) (0.446)Observations 32,722 9603 32,722 9603 32,722 9603 21,338 6810SpecificationSample All Balanced panel All Balanced panel All Balanced panel All Balanced panelTime Trend Flexible Flexible Linear Linear Separate linear Separate linear Industry-specific Industry-specific

Notes: The table presents intensive margin elasticity estimates from Eq. (12) estimated on the period 2006–09, permuting among the combinations of three alternative time trend andbalanced-panel specifications. Standard errors are in parenthesis, which are clustered at the firm level. The treatment group comprises partnership firms, and the results in Panels A and Bare from using sole proprietorships and corporations as the control group. Balanced-panel specifications include firms that report positive earnings in all periods included in theestimation 2006–09. Year fixed effects in Eq. (12) are replaced with a linear time trend in columns (3)–(4), separate linear time trends in columns (5)–(6), and industry-specific trends incolumns (7)–(8). Industry-specific time trend specification includes a complete set of 2-digit industry, year, and industry ×year fixed effects, permitting firms in each industry their ownearnings-growth trend. Placebo results are from the corresponding specification estimated on the period 2006–08, with 2008 assumed as the post-reform period. The estimates areweighted by taxable income so that the elasticity corresponds to the parameter εp in Eq. (10).

Table A4Intensive margin elasticities (2009) — DFL reweighting.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)


(0.106) (0.109) (0.107) (0.106) (0.107) (0.108) (0.111) (0.109) (0.108) (0.109)Placebo 0.050 0.050 0.052 0.050 0.054 0.026 0.028 0.031 0.025 0.034


(0.399) (0.420) (0.396) (0.386) (0.398) (0.332) (0.342) (0.312) (0.317) (0.311)Placebo −0.263 −0.100 −0.007 −0.010 0.292 −0.135 0.005 0.057 −0.033 0.127

(0.657) (0.674) (0.643) (0.634) (0.647) (0.302) (0.321) (0.302) (0.289) (0.310)


68

Observations 12,997 12,997 12,954 12,997 12,954 12,997 12,997 12,954 12,997 12,954ControlsRegion Fixed Effects No Yes No No Yes No Yes No No YesTax Office Fixed Effects No No Yes No Yes No No Yes No YesIndustry Fixed Effects No No No Yes Yes No No No Yes YesTime Trend Flexible Flexible Flexible Flexible Flexible Linear Linear Linear Linear Linear

Notes: The table replicates the results in Table 2 after reweighting the two control samples to match the treatment sample on size and industry dimensions using the DiNardo et al. (1996)method. I group partnerships into 250 (=25 two-digit industries ×10 within-industry size deciles) bins according to the within-industry size-decile distribution of partnerships in 2008.Then within each firm type and year I adjust each bin’s weight so that it carries the same relative weight as the 2008 partnership distribution. Standard errors are in parenthesis, which areclustered at the firm level. The treatment group comprises partnership firms, and the results in Panels A and B are from using sole proprietorships and corporations as the control group.The estimates in column (1) are from the baseline specification; columns (2)–(5) add additional control variables. I do not observe the industry and tax office for all firms, owing to whichthe sample for the corresponding specifications is smaller than that for the others. Columns (6)–(10) replace year fixed effects in Eq. (12) with a linear time trend. Placebo results are fromthe corresponding specification estimated on the period 2006–08, with 2008 assumed as the post-reform period. The estimates are weighted by taxable income so that the elasticitycorresponds to the parameter εp in Eq. (10).

Table A5Intensive margin elasticities (2009) — robustness II.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)


(0.077) (0.078) (0.077) (0.075) (0.075) (0.079) (0.080) (0.080) (0.077) (0.077)Placebo 0.031 0.034 0.036 0.090 0.095 0.029 0.030 0.032 0.074 0.082


(0.273) (0.280) (0.256) (0.264) (0.255) (0.241) (0.247) (0.210) (0.221) (0.203)Placebo −0.222 −0.212 0.051 0.020 0.212 −0.179 −0.120 0.003 −0.094 0.071


Notes: While estimating the intensive margin elasticities reported in all tables other than this, from income shifting concerns I drop firm from the control group whose owners reportincome from a partnership in any of the six periods considered in this study 2006–11. This table replicates the analysis in Table 2 without dropping these firms to show that it is a carefulprecaution only, and that the results with and without these observations are indistinguishable. The treatment group comprises partnership firms, and the results in Panels A and B arefrom using sole proprietorships and corporations as the control group. The estimates in column (1) are from the baseline specification; columns (2)–(5) add additional control variables. Ido not observe the industry and tax office for all firms, owing to which the sample for the corresponding specifications is smaller than that for the others. Columns (6)–(10) replace yearfixed effects in Eq. (12) with a linear time trend. Placebo results are from the corresponding specification estimated on the period 2006–08, with 2008 assumed as the post-reform period.The estimates are weighted by taxable income so that the elasticity corresponds to the parameter εp in Eq. (10).

Table A6Intensive margin elasticities (2009) — time series.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Elasticity 2.233 2.311 2.282 1.958 1.995 2.256 2.262 2.261 2.115 2.127(0.079) (0.083) (0.081) (0.079) (0.080) (0.074) (0.074) (0.074) (0.085) (0.085)

Placebo 0.038 0.049 0.041 0.042 0.047 0.062 0.065 0.064 0.114 0.118(0.044) (0.044) (0.044) (0.049) (0.050) (0.049) (0.049) (0.049) (0.062) (0.062)

Observations 29,489 29,489 29,473 18,135 18,135 9114 9114 9114 6324 6324ControlsRegion Fixed Effects No Yes No No Yes No Yes No No YesTax Office Fixed Effects No No Yes No Yes No No Yes No YesIndustry Fixed Effects No No No Yes Yes No No No Yes YesSample All All All All All Balanced panel Balanced panel Balanced panel Balanced panel Balanced panel

Notes: The table presents intensive margin elasticity estimates from the time-series analog of Eq. (12) estimated on the period 2006–09. Standard errors are in parenthesis, which areclustered at the firm level. The estimates in column (1) are from the baseline specification; columns (2)–(5) add additional control variables. Columns (6)–(10) replicate the results for abalanced-panel sample, which contains firms that report positive earnings in all four periods included in the estimation 2006–09. I do not observe the industry and tax office for all firms,owing to which the sample for the corresponding specifications is smaller than that for the others. Placebo results are from the corresponding specification estimated on the period2006–08, with 2008 assumed as the post-reform period. The estimates are weighted by taxable income so that the elasticity corresponds to the parameter εp in Eq. (10).


69

Table A7Intensive margin elasticities (2009–11) — time series.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Elasticity (2009) 2.233 2.267 2.248 1.951 1.957 1.852 1.859 1.856 1.785 1.791(0.079) (0.082) (0.080) (0.079) (0.079) (0.102) (0.102) (0.102) (0.112) (0.112)

Elasticity (2010) 0.188 0.279 0.225 0.285 0.307 0.465 0.413 0.477 0.426 0.363(0.064) (0.069) (0.065) (0.075) (0.078) (0.088) (0.089) (0.088) (0.099) (0.101)

Elasticity (2011) 0.155 0.293 0.200 0.301 0.357 0.650 0.548 0.665 0.701 0.609(0.063) (0.070) (0.064) (0.076) (0.086) (0.098) (0.097) (0.098) (0.114) (0.115)

Observations 41,267 41,267 41,250 27,509 27,509 5850 5850 5850 4600 4600ControlsRegion Fixed Effects No Yes No No Yes No Yes No No YesTax Office Fixed Effects No No Yes No Yes No No Yes No YesIndustry Fixed Effects No No No Yes Yes Yes No No Yes YesSample All All All All All Balanced panel Balanced panel Balanced panel Balanced panel Balanced panel

Notes: The table presents intensive margin elasticity estimates from the time-series analog of Eq. (12) estimated on the period 2006–11. Standard errors are in parenthesis, which areclustered at the firm level. The estimates in column (1) are from the baseline specification; columns (2)–(5) add additional control variables. Columns (6)–(10) replicate the results for abalanced-panel sample, which contains firms that report positive earnings in all six periods included in the estimation 2006–11. I do not observe the industry and tax office for all firms,owing to which the sample for the corresponding specifications is smaller than that for the others. The estimates are weighted by taxable income so that the elasticity corresponds to theparameter εp in Eq. (10).

Table A8Extensive margin response.

(1) (2) (3) (4) (5)

A: Sole proprietorships as controlPartnership × 2009 −0.182 −0.167 −0.168 −0.173 −0.154

(0.003) (0.003) (0.003) (0.003) (0.003)Partnership × 2010 −0.227 −0.192 −0.204 −0.229 −0.196

(0.003) (0.003) (0.003) (0.004) (0.004)Partnership × 2011 −0.176 −0.140 −0.147 −0.203 −0.166

(0.003) (0.003) (0.003) (0.004) (0.004)Placebo 0.038 0.039 0.039 0.066 0.061

(0.002) (0.002) (0.002) (0.003) (0.003)Observations 3,037,310 3,037,310 2,864,532 954,840 954,822B: Corporations as controlPartnership × 2009 −0.220 −0.195 −0.207 −0.177 −0.150

(0.004) (0.004) (0.004) (0.004) (0.005)Partnership × 2010 −0.250 −0.209 −0.240 −0.205 −0.163

(0.005) (0.005) (0.005) (0.005) (0.005)Partnership × 2011 −0.298 −0.266 −0.283 −0.253 −0.224

(0.005) (0.005) (0.005) (0.005) (0.005)Placebo 0.047 0.038 0.042 0.041 0.027

(0.004) (0.004) (0.004) (0.004) (0.004)Observations 283,575 283,575 283,103 223,616 223,616ControlsRegion Fixed Effects No Yes No No YesTax Office Fixed Effects No No Yes No YesIndustry Fixed Effects No No No Yes YesPrereform Mean 0.579 0.579 0.578 0.531 0.531

Notes: The table presents the results from Eq. (14) estimated on the period 2006–2011 using the linear probability model. Standard errors are in parenthesis, which are clustered at thefirm level. The treatment group comprises partnership firms, and the results in Panels A and B are from using sole proprietorships and corporations as the control group. The sampleincludes all firms that file a return in period t and report earnings in the range zit ∈ [0 650K]. Owing to the tax rule that all registered firms need to file a tax return a large number of firmswith zero earning file return every year. The estimates in column (1) are from the baseline specification; columns (2)–(5) add additional control variables. I do not observe the industryand tax office for a number of firms, owing to which the sample in columns (3)–(5) is smaller than that in the first two columns. The placebo results are from the correspondingspecification estimated on the same period 2006–11 with an additional interaction term Partnership ×2008 included in the regression. The rows titled Placebo show the coefficient andstandard error on this additional interaction term. The last row reports the mean value of the outcome variable in the treatment group in the three prereform periods 2006–08.


70

Table A9Extensive margin response — robustness.

(1) (2) (3) (4) (5) (6)

A: Sole proprietorships as controlPartnership × 2009 −0.182 −0.117 −0.191 −0.095 −0.158 −0.105

(0.003) (0.005) (0.003) (0.005) (0.003) (0.006)Partnership × 2010 −0.227 −0.144 −0.217 −0.127 −0.212 −0.134

(0.003) (0.006) (0.003) (0.006) (0.004) (0.007)Partnership × 2011 −0.176 −0.104 −0.207 −0.144 −0.187 −0.131

(0.003) (0.006) (0.003) (0.006) (0.004) (0.007)Placebo 0.038 0.008 0.077 0.041 0.063 0.017

(0.002) (0.003) (0.002) (0.003) (0.003) (0.004)Observations 3,037,310 1,167,770 3,037,310 1,167,770 954,840 303,602B: Corporations as controlPartnership × 2009 −0.220 −0.158 −0.219 −0.137 −0.157 −0.113

(0.004) (0.009) (0.003) (0.006) (0.005) (0.010)Partnership × 2010 −0.250 −0.189 −0.261 −0.189 −0.188 −0.136

(0.005) (0.011) (0.004) (0.008) (0.005) (0.011)Partnership × 2011 −0.298 −0.252 −0.265 −0.227 −0.237 −0.197

(0.005) (0.011) (0.004) (0.009) (0.006) (0.012)Placebo 0.047 −0.002 0.082 0.041 0.035 −0.013

(0.004) (0.008) (0.002) (0.003) (0.004) (0.008)Observations 283,575 50,401 283,575 50,401 223,616 43,410Sample All Balanced panel All Balanced panel All Balanced panelTime Trend Flexible Flexible Linear Linear Industry-specific Industry-specificPrereform Mean 0.579 0.513 0.579 0.513 0.531 0.491

Notes: The table presents the results from Eq. (14) estimated on the period 2006–2011 using the linear probability model, permuting over the combinations of time-trend and balanced-panel specifications. Standard errors are in parenthesis, which are clustered at the firm level. The treatment group comprises partnership firms, and the results in Panels A and B are fromusing sole proprietorships and corporations as the control group. Balanced-panel specifications include firms that file returns in all periods included in the estimation 2006–11. Year fixedeffects in Eq. (14) are replaced with a linear time trend in columns (3)–(4), and industry-specific trends in columns (5)–(6). Industry-specific time trend specification includes a completeset of 2-digit industry, year, and industry ×year fixed effects, permitting firms in each industry their own earnings-growth trend. The placebo results are from the correspondingspecification estimated on the same period 2006–11 with an additional interaction term Partnerships × 2008 included in the regression. The rows titled Placebo show the coefficient andstandard error on this additional interaction term. The last row reports the mean value of the outcome variable in the treatment group in the three prereform periods 2006–08.

Table A10Extensive margin response — probit.

(1) (2) (3) (4) (5)

A: Sole proprietorships as controlPartnership × 2009 −0.175 −0.138 −0.145 −0.167 −0.139

(0.003) (0.003) (0.003) (0.003) (0.003)Partnership × 2010 −0.204 −0.153 −0.169 −0.210 −0.173

(0.003) (0.003) (0.003) (0.003) (0.004)Partnership × 2011 −0.192 −0.144 −0.150 −0.217 −0.179

(0.003) (0.003) (0.003) (0.004) (0.004)Placebo 0.064 0.061 0.062 0.071 0.064

(0.002) (0.002) (0.002) (0.002) (0.002)Observations 3,037,310 3,037,309 2,864,532 954,840 954,822B: Corporations as controlPartnership × 2009 −0.183 −0.163 −0.173 −0.150 −0.129

(0.002) (0.002) (0.002) (0.003) (0.003)Partnership × 2010 −0.217 −0.204 −0.211 −0.188 −0.172

(0.003) (0.003) (0.003) (0.003) (0.003)Partnership × 2011 −0.221 −0.199 −0.212 −0.196 −0.175

(0.003) (0.003) (0.003) (0.003) (0.004)Placebo 0.078 0.074 0.075 0.073 0.062

(0.002) (0.002) (0.002) (0.002) (0.002)Observations 283,575 283,575 283,103 223,595 223,595ControlsRegion Fixed Effects No Yes No No YesTax Office Fixed Effects No No Yes No YesIndustry Fixed Effects No No No Yes YesPrereform Mean 0.579 0.579 0.578 0.531 0.531

Notes: The table replicates the analysis in Table A8, presenting the results from Eq. (14) estimated on the period 2006–2011 using the probit rather than the linear probability model. Thereported coefficients are the marginal effects, calculated as the difference in probabilities of reporting positive earnings, one with the double-interaction term set equal to one and theother with the interaction term set to zero. Standard errors are in parenthesis, which are clustered at the firm level. The sample includes all firms that file a return in period t and reportearnings in the range zit ∈ [0 650 K]. Owing to the tax rule that all registered firms need to file a tax return a large number of firms with zero earning file return every year. The estimates incolumn (1) are from the baseline specification; columns (2)–(5) add additional control variables. I do not observe the industry and tax office for a number of firms, owing to which thesample in columns (3)–(5) is smaller than that in the first two columns. The placebo results are from the corresponding specification estimated on the same period 2006–11 with anadditional interaction term Partnership ×2008 included in the regression. The rows titled Placebo show the coefficient and standard error on this additional interaction term. The lastrow reports the mean value of the outcome variable in the treatment group in the three prereform periods 2006–08.


71

Table A11

Extensive margin response — sole proprietorships vs. corporations.

(1) (2) (3) (4) (5) (6) (7) (8)

Corporations as controlSole Proprietorship × 2009 −0.022 −0.045 −0.006 −0.037 −0.004 −0.037 −0.001 0.000

(0.003) (0.007) (0.001) (0.001) (0.001) (0.001) (0.003) (0.000)Sole Proprietorship × 2010 −0.108 −0.131 −0.056 −0.114 −0.053 −0.114 −0.058 −0.092

(0.003) (0.007) (0.001) (0.001) (0.001) (0.001) (0.004) (0.008)Placebo −0.033 −0.010 −0.020 0.003 −0.020 0.003 −0.027 −0.013

(0.003) (0.006) (0.001) (0.001) (0.001) (0.001) (0.003) (0.006)Observations 2,916,721 1,132,278 2,916,721 1,132,278 2,916,721 1,132,278 890,868 277,584SpecificationSample All Balanced panel All Balanced panel All Balanced panel All Balanced panelTime Trend Flexible Flexible Linear Linear Separate linear Separate linear Industry-specific Industry-specificPrereform Mean 0.754 0.784 0.754 0.784 0.754 0.784 0.609 0.640

Notes: The table explores if the sole proprietorship extensive margin outcomes are affected by the 2010 tax changes. I report the results from equations similar to Eq. (14) estimated on asample containing sole proprietorships and corporations, using the linear probability model. The top two rows report the coefficients and standard errors on the two difference-in-differences terms, capturing how the sole proprietorship firms’ probability to report positive earnings changed in 2010 and 2011 relative to corporations. The placebo result report thecoefficient and standard error on the additional interaction term Sole Proprietorship × 2009, capturing any preexisting difference in outcomes across the two group of firms. Standarderrors in parenthesis, which have been clustered at the firm level. Balanced-panel specifications include firms that file returns in all periods included in the estimation 2006–11. Year fixedeffects in the baseline specification are replaced with a linear time trend in columns (3)–(4), separate linear time trends in columns (5)–(6), and industry-specific trends in columns(7)–(8). Industry-specific time trend specification includes a complete set of 2-digit industry, year, and industry × year fixed effects, permitting firms in each industry their own earnings-growth trend. The last row reports the mean value of the outcome variable for sole proprietorships in the four pre-2010.

Table A12Income shifting to sole proprietorships.

1(zp,it>0) 1(zs,it>0) 1(zit>0)

(1) (2) (3) (4) (5) (6)A: Sole proprietorships as controlPartnership × 2009 −0.239 −0.245 0.137 0.131 −0.206 −0.212

(0.004) (0.004) (0.003) (0.003) (0.003) (0.003)Partnership × 2010 −0.324 −0.311 0.223 0.236 −0.182 −0.169

(0.004) (0.004) (0.004) (0.004) (0.004) (0.004)Partnership × 2011 −0.292 −0.319 0.306 0.278 −0.157 −0.184

(0.004) (0.004) (0.004) (0.004) (0.004) (0.004)Placebo 0.090 0.129 −0.043 −0.005 0.039 0.078

(0.004) (0.003) (0.003) (0.002) (0.003) (0.003)Observations 2,991,995 2,991,995 2,991,995 2,991,995 2,991,995 2,991,995B: Corporations as controlPartnership × 2009 −0.282 −0.285 0.093 0.088 −0.246 −0.245

(0.005) (0.004) (0.004) (0.003) (0.004) (0.004)Partnership × 2010 −0.358 −0.372 0.187 0.171 −0.213 −0.220

(0.005) (0.004) (0.005) (0.004) (0.005) (0.004)Partnership × 2011 −0.411 −0.401 0.185 0.192 −0.274 −0.252

(0.005) (0.005) (0.005) (0.005) (0.005) (0.005)Placebo 0.141 0.138 −0.028 −0.027 0.088 0.085

(0.004) (0.003) (0.004) (0.003) (0.004) (0.003)Observations 192,604 192,604 192,604 192,604 192,604 192,604Time Trend Flexible Linear Flexible Linear Flexible LinearPrereform Mean 0.701 0.701 0.224 0.224 0.859 0.859

Notes: The table explores income shifting to sole proprietorship business form, by presenting estimates from regressions similar to Eq. (15) using the linear probability model. Thetreatment group comprises partnership owners, defined as individuals who report earnings from a partnership in any of the three prereform periods 2006–08. The results in Panels A andB are from using sole proprietors and corporations as the control group. The three outcome variables are: agent i reports positive partnership earnings (zp) in period t (columns (1)–(2));agent i reports positive sole proprietorship earnings (zs) in period t (columns (3)–(4)); and agent i reports positive overall earnings z=zp+zs in period t (columns (5)–(6)). For the twocontrol groups the outcome variable is coded 1 if firm i reports positive taxable income in period t and zero if it reports zero taxable income. The placebo result report the coefficient andstandard error on the additional interaction term Partnership ×2008, capturing any preexisting difference in outcomes across the treatment and control groups. The last row reports themean value of the outcome variable in the treatment group in the three prereform periods 2006–08. Standard errors are in parenthesis, which have been clustered at the firm level.


72

Table A13Income shifting to sole proprietorships — BP.

1(zp,it>0) 1(zs,it>0) 1(zit>0)

(1) (2) (3) (4) (5) (6)A: Sole proprietorships as controlPartnership × 2009 −0.218 −0.196 0.110 0.131 −0.154 −0.132

(0.006) (0.006) (0.005) (0.005) (0.005) (0.005)Partnership × 2010 −0.278 −0.261 0.195 0.212 −0.119 −0.102

(0.006) (0.006) (0.006) (0.006) (0.005) (0.005)Partnership × 2011 −0.234 −0.275 0.269 0.228 −0.108 −0.148

(0.006) (0.006) (0.006) (0.006) (0.006) (0.006)Placebo 0.060 0.094 −0.042 −0.010 −0.001 0.032

(0.006) (0.006) (0.004) (0.004) (0.004) (0.004)Observations 1,171,375 1,171,375 1,171,375 1,171,375 1,171,375 1,171,375B: Corporations as controlPartnership × 2009 −0.240 −0.233 0.087 0.081 −0.177 −0.157

(0.009) (0.007) (0.008) (0.006) (0.009) (0.006)Partnership × 2010 −0.328 −0.317 0.145 0.137 −0.169 −0.139

(0.009) (0.008) (0.009) (0.007) (0.009) (0.007)Partnership × 2011 −0.370 −0.350 0.133 0.128 −0.244 −0.197

(0.010) (0.010) (0.010) (0.008) (0.010) (0.008)Placebo 0.061 0.110 −0.041 −0.036 −0.000 0.044

(0.009) (0.006) (0.008) (0.004) (0.008) (0.004)Observations 47,147 47,147 47,147 47,147 47,147 47,147Time Trend Flexible Linear Flexible Linear Flexible LinearPrereform Mean 0.612 0.612 0.319 0.319 0.815 0.815

Notes: The table replicates the results in Table A12 on a balanced panel of firms that file return in all periods included in the estimation 2006–11. The treatment group comprisespartnership owners, defined as individuals who report earnings from a partnership in any of the three prereform periods 2006–08. The three outcome variables are: agent i reportspositive partnership earnings (zp) in period t (columns (1)–(2)); agent i reports positive sole proprietorship earnings (zs) in period t (columns (3)–(4)); and agent i reports positive overallearnings z=zp+zs in period t (columns (5)–(6)). For the two control groups the outcome variable is coded 1 if firm i reports positive taxable income in period t and zero if it reports zerotaxable income. The placebo result report the coefficient and standard error on the additional interaction term Partnership ×2008, capturing any preexisting difference in outcomesacross the treatment and control groups. The last row reports the mean value of the outcome variable in the treatment group in the three prereform periods 2006–08. Standard errors arein parenthesis, which have been clustered at the firm level.

Table A14Income shifting to sole proprietorships — robustness.

(1) (2) (3) (4) (5) (6) (7) (8)

A: Sole proprietorships as controlElasticity (2009) −0.238 −0.190 −0.212 −0.211 −0.130 −0.133 −0.190 −0.177

(0.007) (0.011) (0.006) (0.010) (0.010) (0.017) (0.013) (0.020)Elasticity (2010) −31.246 −27.247 −34.370 −30.756 −24.639 −21.234 −24.912 −27.100

(0.545) (0.885) (0.560) (0.911) (1.258) (2.129) (1.027) (1.648)Elasticity (2011) −0.513 −0.452 −0.487 −0.398 −0.337 −0.244 −0.401 −0.410

(0.007) (0.011) (0.007) (0.011) (0.020) (0.034) (0.013) (0.020)Observations 2,991,995 1,171,375 2,991,995 1,171,375 2,991,995 1,171,375 867,567 285,457B: Corporations as controlElasticity (2009) −0.161 −0.151 −0.154 −0.141 −0.130 −0.133 −0.153 −0.116

(0.009) (0.027) (0.007) (0.014) (0.009) (0.018) (0.013) (0.034)Elasticity (2010) −32.717 −25.896 −28.156 −22.443 −24.639 −21.234 −30.994 −23.102

(0.944) (2.766) (0.718) (1.588) (1.154) (2.258) (1.345) (3.456)Elasticity (2011) −0.378 −0.273 −0.396 −0.264 −0.337 −0.244 −0.368 −0.265

(0.011) (0.032) (0.010) (0.024) (0.018) (0.036) (0.016) (0.040)Observations 192,604 47,147 192,604 47,147 192,604 47,147 110,372 21,767SpecificationSample All Balanced panel All Balanced panel All Balanced panel All Balanced panelTime Trend Flexible Flexible Linear Linear Separate linear Separate linear Industry-specific Industry-specific

Notes: The table establishes the robustness the income shifting elasticities in Table 5 by reporting results from additional specifications. Standard errors are in parenthesis, which areclustered at the firm level. The treatment group comprises partnership owners, defined as individuals who report earnings from a partnership in any of the three prereform periods2006–08. The results in Panels A and B are from using sole proprietors and corporations as the control group. Balanced-panel specifications include agents that report positive earnings inall periods included in the estimation 2006–11. Year fixed effects in the baseline specification are replaced with a linear time trend in columns (3)–(4), separate linear time trends incolumns (5)–(6), and industry-specific trends in columns (7)–(8). Industry-specific time trend specification includes a complete set of 2-digit industry, year, and industry ×year fixedeffects, permitting agents in each industry their own earnings-growth trend.


73

Table A15Income shifting to corporations.

(1) (2) (3) (4) (5) (6)

A: Sole proprietorships as controlPartnership × 2009 0.001 0.001 −0.000 −0.002 −0.004 −0.007

(0.001) (0.002) (0.001) (0.002) (0.001) (0.002)Partnership × 2010 0.004 0.004 0.003 0.003 −0.002 −0.004

(0.001) (0.002) (0.001) (0.002) (0.002) (0.003)Partnership × 2011 0.004 0.007 0.004 0.006 −0.002 −0.003

(0.001) (0.002) (0.001) (0.002) (0.002) (0.004)Placebo 0.000 −0.000 0.001 0.003 0.001 0.003

(0.001) (0.002) (0.001) (0.002) (0.001) (0.002)Observations 2,991,995 867,551 2,991,995 867,551 2,991,995 867,551B: Corporations as controlPartnership × 2009 0.001 0.001 0.000 −0.000 −0.004 −0.006

(0.001) (0.002) (0.001) (0.002) (0.001) (0.002)Partnership × 2010 0.005 0.006 0.004 0.005 −0.002 −0.004

(0.001) (0.002) (0.001) (0.002) (0.002) (0.003)Partnership × 2011 0.006 0.008 0.006 0.009 −0.002 −0.003

(0.001) (0.002) (0.001) (0.002) (0.002) (0.004)Placebo 0.004 0.005 0.002 0.003 0.001 0.002

(0.001) (0.002) (0.001) (0.002) (0.001) (0.002)Observations 192,595 110,363 192,595 110,363 192,595 110,363ControlsRegion Fixed Effects No Yes No Yes No YesTax Office Fixed Effects No Yes No Yes No YesIndustry Fixed Effects No Yes No Yes No YesTime Trend Flexible Flexible Linear Linear Separate linear Separate linearPrereform Mean 0.021 0.027 0.021 0.027 0.021 0.027

Notes: The table explores income shifting to the corporate business form. The estimates are from the difference-in-differences regression Eq. (15) with an indicator showing if i reportspositive dividend income in period t as the outcome variable. The treatment group comprises partnership owners, defined as individuals who report earnings from a partnership in any ofthe three prereform periods 2006–08. The results in Panels A and B are from using sole proprietors and corporations as the control group. The placebo result report the coefficient andstandard error on the additional interaction term Partnership ×2008, capturing any preexisting difference in outcomes across the treatment and control groups. The last row reports themean value of the outcome variable in the treatment group in the three prereform periods 2006–08. Standard errors are in parenthesis, which have been clustered at the firm level.

Table A16Income shifting to corporations — BP.

(1) (2) (3) (4) (5) (6)

A: Sole proprietorships as controlPartnership × 2009 −0.001 −0.003 −0.002 −0.004 −0.005 −0.005

(0.001) (0.002) (0.001) (0.002) (0.002) (0.004)Partnership × 2010 −0.000 0.001 0.000 0.001 −0.004 −0.001

(0.001) (0.002) (0.001) (0.002) (0.003) (0.005)Partnership × 2011 0.003 0.004 0.002 0.004 −0.003 0.002

(0.001) (0.003) (0.001) (0.003) (0.003) (0.006)Placebo 0.001 0.001 0.002 0.002 0.002 0.003

(0.001) (0.003) (0.001) (0.002) (0.001) (0.002)Observations 1,171,375 285,457 1,171,375 285,457 1,171,375 285,457B: Corporations as controlPartnership × 2009 −0.002 −0.002 −0.003 −0.003 −0.005 −0.005

(0.001) (0.002) (0.001) (0.002) (0.002) (0.004)Partnership × 2010 0.001 0.002 −0.001 0.002 −0.004 −0.002

(0.001) (0.002) (0.001) (0.002) (0.003) (0.005)Partnership × 2011 0.003 0.006 0.001 0.005 −0.003 0.000

(0.001) (0.002) (0.002) (0.002) (0.003) (0.006)Placebo 0.002 0.003 0.002 0.003 0.002 0.003

(0.001) (0.002) (0.001) (0.002) (0.001) (0.002)Observations 47,153 21,773 47,153 21,773 47,153 21,773ControlsRegion Fixed Effects No Yes No Yes No YesTax Office Fixed Effects No Yes No Yes No YesIndustry Fixed Effects No Yes No Yes No YesTime Trend Flexible Flexible Linear Linear Separate linear Separate linearPrereform Mean 0.028 0.031 0.028 0.031 0.028 0.031

Notes: The table replicates the results in Table A15 on a balanced panel sample containing agents that report positive taxable earnings in all periods included in the estimation2006–2011. The estimates are from the difference-in-differences regression Eq. (15) with an indicator showing if i reports positive dividend income in period t as the outcome variable.The treatment group comprises partnership owners, defined as individuals who report earnings from a partnership in any of the three prereform periods 2006–08. The results in Panels Aand B are from using sole proprietors and corporations as the control group. The placebo result report the coefficient and standard error on the additional interaction term Partnership×2008, capturing any preexisting difference in outcomes across the treatment and control groups. The last row reports the mean value of the outcome variable in the treatment group inthe three prereform periods 2006–08. Standard errors are in parenthesis, which have been clustered at the firm level.


74

Table A17Heterogeneity in intensive margin response — firm characteristics.

(1) (2) (3) (4) (5) (6) (7) (8)

Partnership × 2009 2.233 2.573 2.709 2.247 2.984 2.439 2.608 3.538(0.077) (0.243) (0.114) (0.082) (0.099) (0.082) (0.095) (0.292)

Partnership × 2009× Electronic Return Filer −0.565 0.055

(0.250) (0.284)× Firm Size −1.670 −0.730

(0.157) (0.178)× Firm Age −0.642 −0.432

(0.184) (0.231)× Tax Withholding −2.260 −1.906

(0.134) (0.154)× Withholding Agent −1.634 −0.195

(0.170) (0.232)× VAT-Registered −1.498 −0.927

(0.137) (0.172)Observations 848,466 848,466 620,735 811,064 848,466 848,466 848,466 613,116

Notes: The table explores heterogeneity in intensive margin elasticities across firms with different characteristics. I report the results from triple-difference version of Eq. (12), includingthe firm characteristic and triple-interaction (Partnership × Year × Firm Characteristic) dummies into the regression. The estimates in column (1) report the income-weighted averageelasticity; the estimates in the subsequent columns break down the elasticity by firm-characteristic indicated in each row. Details of the firm characteristics variables are given inAppendix A.1. Standard errors are in parenthesis, which are clustered at the firm level.

Table A18Heterogeneity in intensive margin response — income groups.

Taxable Income ≤ (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Sole proprietorships as control250,000 3.134 3.167 3.155 2.846 2.865 3.106 3.140 3.129 2.821 2.839

(0.117) (0.118) (0.117) (0.113) (0.113) (0.117) (0.119) (0.118) (0.113) (0.114)[795,997] [795,997] [759,265] [146,081] [146,077] [795,997] [795,997] [759,265] [146,081] [146,073]

350,000 2.829 2.856 2.846 2.585 2.597 2.808 2.838 2.826 2.558 2.566(0.097) (0.098) (0.097) (0.092) (0.092) (0.098) (0.099) (0.098) (0.092) (0.093)[828,288] [828,288] [791,125] [161,555] [161,551] [828,288] [828,288] [791,125] [161,555] [161,542]

450,000 2.542 2.567 2.558 2.311 2.323 2.526 2.553 2.542 2.286 2.293(0.083) (0.084) (0.083) (0.079) (0.079) (0.084) (0.085) (0.085) (0.080) (0.080)[840,398] [840,398] [803,086] [168,869] [168,864] [840,398] [840,398] [803,086] [168,869] [168,852]

550,000 2.338 2.360 2.356 2.116 2.128 2.331 2.354 2.348 2.096 2.105(0.076) (0.077) (0.076) (0.075) (0.075) (0.078) (0.079) (0.078) (0.076) (0.077)[846,310] [846,310] [808,945] [172,930] [172,925] [846,310] [846,310] [808,945] [172,930] [172,909]

650,000 2.233 2.253 2.251 1.999 2.009 2.219 2.241 2.238 1.973 1.981(0.077) (0.078) (0.077) (0.074) (0.075) (0.079) (0.080) (0.080) (0.077) (0.077)[848,466] [848,466] [811,075] [174,475] [174,470] [848,466] [848,466] [811,075] [174,475] [174,450]

ControlsRegion Fixed Effects No Yes No No Yes No Yes No No YesTax Office Fixed Effects No No Yes No Yes No No Yes No YesIndustry Fixed Effects No No No Yes Yes No No No Yes YesTime Trend Flexible Flexible Flexible Flexible Flexible Linear Linear Linear Linear Linear

Notes: The table explores heterogeneity in intensive margin response across high- vs. low-income firms. I report the results from Eq. (12), estimated on the period 2006–09. Treatment andcontrol groups are partnerships and sole proprietorships respectively. Each row of the table reports results from the regression, restricting the sample to firms that have base periodincome up to the limit −z indicated in the row i.e. ∈ −z z(0 ]it . Standard errors are in parenthesis, which have been clustered at the firm level. Number of observations in each regression aregiven in square brackets.

Table A19Heterogeneity in extensive margin response — firm characteristics.

(1) (2) (3) (4) (5) (6) (7) (8)

Partnership × 2009 −0.182 −0.385 −0.341 −0.202 −0.331 −0.207 −0.267 −0.494(0.003) (0.006) (0.005) (0.003) (0.004) (0.003) (0.003) (0.011)

× Electronic Return Filer 0.277 0.088(0.006) (0.012)

× Firm Size 0.232 0.153(0.010) (0.011)

× Firm Age 0.095 0.061(0.006) (0.011)

× Tax Withholding 0.191 0.115(0.008) (0.010)


75

× Withholding Agent 0.254 0.087(0.009) (0.014)

× VAT-Registered 0.240 0.126(0.005) (0.010)

Observations 2,047,399 2,047,399 1,050,675 1,931,287 1,519,360 2,047,399 2,047,399 1,024,080

Notes: The table explores heterogeneity in extensive margin response across firms with different characteristics. I report the results from the following triple-difference version of Eq. (14)estimated on the period 2006–09

> = + + + × + × × + +z α β Partnership β Xtc β Partnership β Partnership Xtc λ u1( 0) 2009 2009 ,it i i it it t it0 1 2 3

where Xtci is the firm-characteristic variable. The estimates in column (1) capture the average, tax-driven reduction in the propensity to report positive earnings by partnership firms. Theestimates in the subsequent columns break down the response by firm-characteristic indicated in each row: top rows of the columns report β2 and the bottom rows β3 from the aboveregressions. Details of the firm characteristics variables are given in Appendix A.1. Standard errors are in parenthesis, which are clustered at the firm level.

Table A20Taxable income responses (partnerships vs. sole proprietorships).

Notch point Earnings response Structural elasticity

Partnerships Sole proprietorships Difference Partnerships Sole proprietorships Difference(1) (2) (3) (4) (5) (6) (7)200 K 15,500 18,000 −2500 0.229 0.308 −0.079

(3283) (5141) (6099) (0.102) (0.207) (0.231)300 K 24,000 29,000 −5000 0.080 0.121 −0.041

(4468) (5941) (7434) (0.036) (0.053) (0.064)400 K 24,500 35,500 −11,000 0.041 0.095 −0.054

(4075) (7688) (8702) (0.016) (0.044) (0.047)500 K 32,500 34,500 −2000 0.045 0.052 −0.007

(7665) (5894) (9669) (0.027) (0.020) (0.034)600 K 22,500 30,000 −7500 0.008 0.028 −0.020

(5956) (5342) (8000) (0.010) (0.009) (0.013)

Notes: This table illustrates how partnerships and sole proprietorships compare in terms of their tax behavior. The table partially recreates Table II in Kleven and Waseem (2013) , andestimates earnings response and taxable income elasticity separately for sole proprietorships and partnerships, exploiting their bunching responses to the notches in the baseline taxsystem. Standard errors are in parentheses. The results show that though earnings responses and elasticities are relatively smaller for partnerships, the differences are not significant eitherstatistically or economically.

Table A21Industry distribution of firms.

Industry label Industry description

(1) (2)1 Personal service activities including washing and dry-cleaning, hairdressing and other beauty treatment, funeral services2 Non-specialized wholesale trade3 Other manufacturing not elsewhere classified4 Retail sale in non-specialized stores5 Spinning, weaving, and finishing of textile products6 Manufacture of knitted, crocheted, and other fabrics7 Activities of business and employers membership8 Retail sale of clothing, footwear and leather articles, pharmaceutical, and cosmetic goods in specialized stores9 Motor repair services10 Wholesale of construction materials, hardware, metals, and metal ores11 Manufacture of prepared meals, chocolate, and sugar confectionary12 Wholesale of machinery, equipment, and supplies13 Growing of cereals (except rice), leguminous crops, and oil seeds14 Wholesale of textiles, clothing, footwear, and other household goods15 Construction of buildings16 Retail sale of electrical household appliances, furniture, lighting equipment in specialized stores17 Business support service activities including packaging18 Sale of motor vehicles19 Real estate activities with own or leased property20 Wholesale of food, beverages, and tobacco21 Manufacture of gas; distribution of gaseous fuels through mains22 Manufacture of grain mill and starch products23 Manufacture of sports goods24 Manufacture of medical and dental instruments and supplies25 Retail sale in specialized stores

Notes: This table presents the detailed description of the 25 industries shown in the two top panels of Fig. A1. Column (1) corresponds to the industry label shown along the x-axis of theplot. Column (2) provides the detailed description of the industry.


76

References

Andreoni, J., Erard, B., Feinstein, J., 1998. Tax complicance. J. Econ. Lit. 36, 818–860.Bachas, P., Soto, M., 2017. Not(ch) your Average Tax System: Corporate Taxation Under Weak

Enforcement. Working Paper.Best, M.C., Brockmeyer, A., Kleven, H.J., Spinnewijn, J., Waseem, M., 2015. Production versus

revenue efficiency with limited tax capacity: theory and evidence from Pakistan. J. Polit.Econ. 123 (6), 1311–1355.

Bruhn, M., McKenzie, D., 2014. Entry regulation and the formalization of microenterprises indeveloping countries. World Bank Res. Obs. 29 (2), 186–201.

Carrillo, P., Pomeranz, D., Singhal, M., 2017. Dodging the taxman: firm misreporting and limitsto tax enforcement. Am. Econ. J. Appl. Econ. 9 (2), 144–164.

Chetty, R., 2009a. Is the taxable income elasticity sufficient to calculate deadweight loss? Theimplications of evasion and avoidance. Am. Econ. J. Econ. Pol. 1 (2), 31–52.

Chetty, R., 2009b. Sufficient statistics for welfare analysis: a bridge between structural andreduced-form methods. Annu. Rev. Econ. 1 (1), 451–488.

Cooper, M., McClelland, J., Pearce, J., Prisinzano, R., Sullivan, J., Yagan, D., Zidar, O., Zwick,E., 2015. Business in the United States: Who Owns It and How Much Tax Do They Pay?Working Paper 21651. National Bureau of Economic Research October.

DiNardo, J., Fortin, N.M., Lemieux, T., 1996. Labor market institutions and the distribution ofwages, 1973–1992: a semiparametric approach. Econometrica 64 (5), 1001–1044.

Feldstein, M., 1999. Tax avoidance and the deadweight loss of the income tax. Rev. Econ. Stat.81 (4), 674–680.

Fisman, R., Wei, S.-J., et al., 2004. Tax rates and tax evasion: evidence from “missing imports”in China. J. Polit. Econ. 112 (2), 471.

Fuest, C., Riedel, N., 2009. Tax Evasion, Tax Avoidance and Tax Expenditures in DevelopingCountries: A Review of the Literature. Report prepared for the UK Department forInternational Development (DFID).

Goolsbee, A., 2004. The impact of the corporate income tax: evidence from state organizational

form data. J. Public Econ. 88, 2283–2299.Gordon, R., Li, W., 2009. Tax structures in developing countries: many puzzles and a possible

explanation. J. Public Econ. 93 (7-8), 855–866.Gordon, R., Slemrod, J., 2000. Are ‘real’ responses to taxes simply income shifting between

corporate and personal tax bases? In: Slemrod, J. (Ed.), Does atlas shrug? The economicsconsequences of taxing the rich.

Gordon, R.H., MacKie-Mason, J.K., 1997. How much do taxes discourage incorporation? J.Financ. 52 (2), 477–505.

Kleven, H.J., Knudsen, M.B., Kreiner, C.T., Pedersen, S., Saez, E., 2011, 05. Unwilling or unableto cheat? Evidence from a tax audit experiment in Denmark. Econometrica 79 (3),651–692.

Kleven, H.J., Waseem, M., 2013. Using notches to uncover optimization frictions and structuralelasticities: theory and evidence from Pakistan. Q. J. Econ. 128, 669–723.

Kleven, H.J., Kreiner, C.T., Saez, E., 2016. Why can modern governments tax so much? anagency model of firms as fiscal intermediaries. Economica 83 (330), 219–246.

Kopczuk, W., 2012. The Polish Business “Flat” Tax and its Effect on Reported Incomes: A ParetoImproving Tax Reform? Columbia University Working Paper.

Marion, J., Muehlegger, E., 2008. Measuring illegal activity and the effects of regulatory in-novation: tax evasion and the dyeing of untaxed diesel. J. Polit. Econ. 116 (4), 633–666.

Nevo, A., 2000. A practitioner’s guide to estimation of random-coefficients logit models ofdemand. J. Econ. Manag. Strateg. 9 (4), 513–548.

Saez, E., Slemrod, J., Giertz, S., 2012. The elasticity of taxable income with respect to marginaltax rates: a critical review. J. Econ. Lit. 50, 3–50.

Slemrod, J., Kopczuk, W., 2002. The optimal elasticity of taxable income. J. Public Econ. 84 (1),91–112.

Slemrod, J., Yitzhaki, S., 2002. Tax avoidance, evasion, and administration. In: Auerbach, A.J.,Feldstein, M. (Eds.), Handbook of Public Economics. vol. 3.

Waseem, M., 2017. Using the Movement of Exemption Cutoff to Estimate Tax Evasion: Evidencefrom Pakistan. University of Manchester Working Paper.


77

http://refhub.elsevier.com/S0047-2727(17)30187-1/rf0005
























































Documents

Journal of Public Economics - Mazhar Waseemmazharwaseem.com/static/uploads/Waseem2018_JPubE... · (IGC), Pakistan Program is gratefully acknowledged. E-mail address: [email protected]